34 253 FEDERAL REPORTER, 3d SERIES - Berkeley Law · 34 253 FEDERAL REPORTER, 3d SERIES UNITED STATES of America, Appellee, v. MICROSOFT CORPORATION, Appellant. Nos. 00–5212 and

34 253 FEDERAL REPORTER, 3d SERIES

UNITED STATES of America,Appellee,

v.

MICROSOFT CORPORATION,Appellant.

Nos. 00–5212 and 00–5213.

United States Court of Appeals,District of Columbia Circuit.

Argued Feb. 26 and 27, 2001.

Decided June 28, 2001.

Rehearing Denied Aug. 2, 2001.

United States and individual statesbrought antitrust action against manufac-turer of personal computer operating sys-tem and Internet web browser. The Unit-ed States District Court for the Districtof Columbia, Thomas Penfield Jackson, J.,concluded that manufacturer had commit-ted monopolization, attempted monopoliza-tion, and tying violations of the ShermanAct, 87 F.Supp.2d 30, and issued remedialorder requiring manufacturer to submitproposed plan of divestiture, 97 F.Supp.2d59. Manufacturer appealed, and states pe-titioned for certiorari. The Supreme Courtdeclined to hear direct appeal, denied pe-tition, and remanded, 530 U.S. 1301, 121S.Ct. 25, 147 L.Ed.2d 1048. The Court ofAppeals held that: (1) manufacturer com-mitted monopolization violation; (2) manu-facturer did not commit attempted mo-nopolization violation; (3) rule of reason,rather than per se analysis, applied totying claim; (4) remand was required todetermine if manufacturer committed ty-ing violation; (5) vacation of remedies de-cree was required; and (6) district judge’scomments to the press while the case waspending required his disqualification onremand.

Affirmed in part, reversed in part, andremanded in part.

1. Monopolies O12(1.3)

Offense of monopolization under Sher-man Act has two elements: (1) possessionof monopoly power in relevant market and(2) willful acquisition or maintenance ofthat power as distinguished from growthor development as a consequence of a su-perior product, business acumen, or histor-ic accident. Sherman Act, § 2, as amend-ed, 15 U.S.C.A. § 2.

2. Federal Courts O776

Court of Appeals reviews legal ques-tions de novo.


While merely possessing monopolypower is not itself an antitrust violation, itis a necessary element of a monopolizationcharge. Sherman Act, § 2, as amended,15 U.S.C.A. § 2.


Firm is a ‘‘monopolist’’ if it can profit-ably raise prices substantially above thecompetitive level. Sherman Act, § 2, asamended, 15 U.S.C.A. § 2.

See publication Words and Phras-es for other judicial constructionsand definitions.


Monopoly power may be inferred froma firm’s possession of a dominant share ofa relevant market that is protected byentry barriers; ‘‘entry barriers’’ are fac-tors, such as certain regulatory require-ments, that prevent new rivals from timelyresponding to an increase in price abovethe competitive level. Sherman Act, § 2,as amended, 15 U.S.C.A. § 2.



Because the ability of consumers toturn to other suppliers restrains a firm

35U.S. v. MICROSOFT CORP.Cite as 253 F.3d 34 (D.C. Cir. 2001)

from raising prices above competitive lev-el, the relevant market in monopolizationcase must include all products reasonablyinterchangeable by consumers for thesame purposes. Sherman Act, § 2, asamended, 15 U.S.C.A. § 2.

7. Federal Courts O763.1Court of Appeals would adhere to

district court’s decision to exclude non-Intel compatible operating systems, andoperating systems for non-PC devicesfrom relevant market for purposes of mo-nopolization claim against manufacturer ofpersonal computer (PC) operating system,where manufacturer did not point to evi-dence contradicting district court’s find-ings or allege that supporting record evi-dence was insufficient. Sherman Act, § 2,as amended, 15 U.S.C.A. § 2.

8. Monopolies O12(1.3)Relevant market for monopolization

claim against manufacturer of personalcomputer operating system did not include‘‘middleware’’ products that could eventu-ally take over operating system functions;consumers could not abandon their operat-ing systems and switch to middleware inresponse to sustained price for manufac-turer’s operating system above competitivelevel, and it was unlikely that middlewarewould overtake manufacturer’s operatingsystem as primary platform for softwaredevelopment in near future. ShermanAct, § 2, as amended, 15 U.S.C.A. § 2.

9. Monopolies O28(7.5)Evidence that manufacturer of operat-

ing system for personal computers hadmore than 95% market share and thatconsumer preference for operating sys-tems for which substantial number of ap-plications had been written and softwaredeveloper preference for operating sys-tems with substantial consumer base cre-ated barrier to entry into operating systemmarket was sufficient to support finding

that manufacturer possessed monopolypower, as required to support monopoliza-tion claim. Sherman Act, § 2, as amend-ed, 15 U.S.C.A. § 2.


Although existence of monopoly powerordinarily may be inferred from the pre-dominant share of the market, because ofthe possibility of competition from newentrants, looking to current market sharealone can be misleading when evaluatingmonopolization claim. Sherman Act, § 2,as amended, 15 U.S.C.A. § 2.


Direct proof of monopoly power wasnot required to support monopolizationclaim against manufacturer of operatingsystem for personal computers, even as-suming that software market was uniquelydynamic in the long term, where noprompt substitutes for manufacturer’s op-erating system were available. ShermanAct, § 2, as amended, 15 U.S.C.A. § 2.


A firm violates Sherman Act’s monop-olization provision only when it acquires ormaintains, or attempts to acquire or main-tain, a monopoly by engaging in exclusion-ary conduct as distinguished from growthor development as a consequence of a su-perior product, business acumen, or histor-ic accident. Sherman Act, § 2, as amend-ed, 15 U.S.C.A. § 2.


To be condemned as exclusionary, amonopolist’s act must have an ‘‘anticom-petitive effect,’’ that is, it must harm thecompetitive process and thereby harm con-sumers. Sherman Act, § 2, as amended,15 U.S.C.A. § 2.



14. Monopolies O12(1.3), 28(1.4)Harm to one or more competitors will

not suffice to establish anticompetitive ef-fect under Sherman Act’s monopolizationprovision. Sherman Act, § 2, as amended,15 U.S.C.A. § 2.

15. Monopolies O12(1.3)Sherman Act directs itself not against

conduct which is competitive, even severe-ly so, but against conduct which unfairlytends to destroy competition itself. Sher-man Act, § 2, as amended, 15 U.S.C.A.§ 2.

16. Monopolies O28(7.2)Plaintiff in monopolization case, on

whom the burden of proof rests, mustdemonstrate that the monopolist’s conductindeed has the requisite anticompetitiveeffect. Sherman Act, § 2, as amended, 15U.S.C.A. § 2.

17. Monopolies O28(1.4)In monopolization case brought by a

private plaintiff, the plaintiff must showthat its injury is of the type that thestatute was intended to forestall; no less ina case brought by the Government, it mustdemonstrate that the monopolist’s conductharmed competition, not just a competitor.Sherman Act, § 2, as amended, 15U.S.C.A. § 2.

18. Monopolies O28(7.2)If a plaintiff successfully establishes a

prima facie case under Sherman Act’s mo-nopolization provision by demonstratinganticompetitive effect, then the monopolistmay proffer a procompetitive justificationfor its conduct. Sherman Act, § 2, asamended, 15 U.S.C.A. § 2.

19. Monopolies O28(7.1)If monopolist asserts a procompetitive

justification for its conduct, a nonpretextu-al claim that its conduct is indeed a form ofcompetition on the merits because it in-

volves, for example, greater efficiency orenhanced consumer appeal, then the bur-den shifts back to the plaintiff to rebutthat claim. Sherman Act, § 2, as amend-ed, 15 U.S.C.A. § 2.


If monopolist’s procompetitive justifi-cation for its conduct stands unrebutted,then plaintiff in monopolization case mustdemonstrate that the anticompetitive harmof the conduct outweighs the procompeti-tive benefit. Sherman Act, § 2, as amend-ed, 15 U.S.C.A. § 2.


In considering whether monopolist’sconduct on balance harms competition andis therefore condemned as exclusionary forpurposes of Sherman Act’s monopolizationprovision, court’s focus is upon the effectof that conduct, not upon the intent behindit; evidence of the intent behind the con-duct of a monopolist is relevant only to theextent it helps courts understand the likelyeffect of the monopolist’s conduct.


Licensing restrictions imposed onoriginal equipment manufacturers by man-ufacturer of operating system for personalcomputers, including the prohibition of theremoval of desktop icons, folders, andStart menu entries, and modifications toinitial boot sequence, had anticompetitiveeffect, supporting monopolization claim; re-strictions prevented promotion of multipleInternet access providers and browsersand reduced rival browsers’ usage share.Sherman Act, § 2, as amended, 15U.S.C.A. § 2.

23. Monopolies O12(5, 15)

Intellectual property rights do notconfer a privilege to violate the antitrustlaws.



License restriction imposed on origi-nal equipment manufacturers by manufac-turer of operating system for personalcomputers which prohibited automaticallylaunching a substitute user interface uponcompletion of the boot process was neces-sary to prevent substantial alteration ofoperating system manufacturer’s copy-righted work, and was not an exclusionarypractice that violated Sherman Act’s mo-nopolization provision. Sherman Act, § 2,as amended, 15 U.S.C.A. § 2.


Evidence was insufficient to establishthat licensing restrictions imposed onoriginal equipment manufacturers bymanufacturer of operating system for per-sonal computers, including altering theappearance of the desktop or promotingprograms in the boot sequence, were nec-essary to prevent actions that would sub-stantially reduce value of its copyrightedwork, as defense to monopolization claim.Sherman Act, § 2, as amended, 15U.S.C.A. § 2.


Licensing restrictions imposed onoriginal equipment manufacturers by man-ufacturer of operating system for personalcomputers which made it more difficult todistribute competing Internet browser vio-lated Sherman Act’s monopolization provi-sion, even though restrictions did not com-pletely bar competitor from distributing itsbrowser, where restrictions barred rivalsfrom all cost-efficient means of distribu-tion. Sherman Act, § 2, as amended, 15U.S.C.A. § 2.


Judicial deference to product innova-tion does not mean that a monopolist’sproduct design decisions are per se lawful.


Manufacturer’s integration of its In-ternet browser into its operating systemfor personal computers had an anticompet-itive effect, for purposes of Sherman Act’smonopolization provision; excluding manu-facturer’s browser from operating system’s‘‘Add/Remove Programs’’ utility had effectof significantly reducing usage of rivals’products, system’s override feature de-terred consumers from using rival brow-sers, and its commingling of browsing andnon-browsing code deterred original equip-ment manufacturers from pre-installing ri-val browsers. Sherman Act, § 2, asamended, 15 U.S.C.A. § 2.


Evidence in antitrust action was suffi-cient to support finding that manufacturerof operating system for personal comput-ers placed code specific to Web browsingin the same files as code that providedoperating system functions, prohibitingoriginal equipment manufacturers from re-moving manufacturer’s Internet browser;government expert testified that manufac-turer designed its browser so that some ofthe code it used co-resided in same libraryfiles as other code needed for the operat-ing system.


Manufacturer’s exclusion of its Inter-net browser from the ‘‘Add/Remove Pro-grams’’ utility in its personal computeroperating system and its commingling ofbrowser and operating system code wasexclusionary conduct, in violation of Sher-man Act’s monopolization provision; suchactions increased manufacturer’s browserusage share, and manufacturer failed toshow that its conduct served a purposeother than protecting its operating systemmonopoly. Sherman Act, § 2, as amend-ed, 15 U.S.C.A. § 2.


31. Monopolies O12(1.8)Manufacturer of operating system for

personal computers did not violate Sher-man Act’s monopolization provision by de-veloping software package that allowed In-ternet access providers to customize titlebar for manufacturer’s Internet browserand offering the package to providers freeof charge. Sherman Act, § 2, as amended,15 U.S.C.A. § 2.

32. Monopolies O12(1.4)Antitrust laws do not condemn even a

monopolist for offering its product at anattractive price.

33. Monopolies O17(2.2)Personal computer operating system

manufacturer’s exclusive dealing agree-ments with Internet access providers, un-der which providers agreed not to promoteInternet browsers of manufacturer’s com-petitors, violated Sherman Act’s monopoli-zation provision; by ensuring that majorityof all providers’ subscribers were offeredmanufacturer’s browser either as defaultbrowser or as the only browser, manufac-turer’s deals with the providers had signif-icant effect in preserving manufacturer’soperating system monopoly. ShermanAct, § 2, as amended, 15 U.S.C.A. § 2.

34. Monopolies O17(2.2)When exclusive deal is challenged in

antitrust action, plaintiff must both definerelevant market and prove the degree offoreclosure.

35. Monopolies O17(2.2)Monopolist’s use of exclusive con-

tracts, in certain circumstances, may giverise to a monopolization violation underSherman Act even though the contractsforeclose less than the roughly 40% or 50%share usually required in order to establisha restraint of trade violation. ShermanAct, §§ 1, 2, as amended, 15 U.S.C.A.§§ 1, 2.

36. Monopolies O17.5(1)

Manufacturer of operating systemfor personal computers did not violateSherman Act’s monopolization provisionby granting Internet content providersfree licenses to bundle manufacturer’s In-ternet browser and offering inducementsnot to offer competitor’s browser, absentevidence that such restrictions had sub-stantial deleterious impact on competi-tor’s usage share. Sherman Act, § 2, asamended, 15 U.S.C.A. § 2.


Agreements between manufacturer ofoperating system for personal computersand independent software vendors, underwhich developers agreed to use manufac-turer’s Internet browser as default brows-ing software for any software they devel-oped with hypertext-based user interface,violated Sherman Act’s monopolizationprovision; by keeping rival browsers fromgaining widespread distribution, the dealshad a substantial effect in preserving man-ufacturer’s operating system monopoly,and manufacturer offered no procompeti-tive justification. Sherman Act, § 2, asamended, 15 U.S.C.A. § 2.


Personal computer operating systemmanufacturer’s exclusive dealing arrange-ment with computer company, under whichoperating system manufacturer agreed torelease up-to-date versions of businessproductivity software compatible with com-pany’s computers in exchange for compa-ny’s agreement to make manufacturer’sInternet browser the default browser onits computers, violated Sherman Act’s mo-nopolization provision; arrangement hadsubstantial effect in restricting distributionof rival browsers, and manufacturer of-fered no procompetitive justification.


Sherman Act, § 2, as amended, 15U.S.C.A. § 2.

39. Monopolies O12(1.3)Manufacturer’s development and pro-

motion of a ‘‘Java Virtual Machine’’ (JVM)which translated bytecode into instructionsfor manufacturer’s personal computer op-erating system and allowed Java applica-tions to run faster on its operating systemthan competitor’s JVM did not violateSherman Act’s monopolization provision;although manufacturer’s JVM was incom-patible with competitor’s product, it al-lowed applications to run more swiftly anddid not itself have any anticompetitive ef-fect. Sherman Act, § 2, as amended, 15U.S.C.A. § 2.


manufacturer’s contracts with independentsoftware vendors, which required use ofmanufacturer’s ‘‘Java Virtual Machine’’(JVM) as default in their Java applications,were exclusionary, in violation of ShermanAct’s monopolization provision; agreementswere anticompetitive because they fore-closed a substantial portion of the field forJVM distribution and, in so doing, protect-ed manufacturer’s operating system mo-nopoly from a middleware threat, andmanufacturer offered no procompetitivejustification. Sherman Act, § 2, asamended, 15 U.S.C.A. § 2.


manufacturer’s deception of Java develop-ers regarding the operating system-specif-ic nature of software development tools itcreated to assist independent softwarevendors in designing Java applications wasexclusionary conduct, in violation of Sher-man Act’s monopolization provision; manu-facturer’s conduct served to protect itsmonopoly in its operating system in man-ner not attributable either to superiority of

the operating system or the acumen of itsmakers. Sherman Act, § 2, as amended,15 U.S.C.A. § 2.


manufacturer’s threat to retaliate againstmicroprocessor manufacturer if it did notstop developing a fast, cross-platform‘‘Java Virtual Machine’’ (JVM) was exclu-sionary conduct, in violation of ShermanAct’s monopolization provision; develop-ment of such a JVM would have threat-ened operating system manufacturer’s mo-nopoly in operating system market, andmanufacturer offered on procompetitivejustification for its conduct. Sherman Act,§ 2, as amended, 15 U.S.C.A. § 2.


manufacturer could not be held liable un-der Sherman Act’s monopolization provi-sion based on its general ‘‘course of con-duct,’’ where specific acts by manufacturerthat harmed competition were not identi-fied. Sherman Act, § 2, as amended, 15U.S.C.A. § 2.

44. Monopolies O24(7.1)Causal link between maintenance of

manufacturer’s personal computer operat-ing system monopoly and its anticompeti-tive conduct in foreclosing distributionchannels for rival Internet browser andJava technologies was not required to holdmanufacturer liable for monopolization vio-lation in action seeking injunctive relief.Sherman Act, § 2, as amended, 15U.S.C.A. § 2.

45. Monopolies O12(1.3)To establish a Sherman Act violation

for attempted monopolization, a plaintiffmust prove (1) that the defendant hasengaged in predatory or anticompetitiveconduct with (2) a specific intent to monop-olize and (3) a dangerous probability of


achieving monopoly power. Sherman Act,§ 2, as amended, 15 U.S.C.A. § 2.


Because the Sherman Act does notidentify the activities that constitute theoffense of attempted monopolization, thecourt must examine the facts of each case,mindful that the determination of whatconstitutes an attempt is a question ofproximity and degree. Sherman Act, § 2,as amended, 15 U.S.C.A. § 2.

47. Monopolies O12(1.3)Manufacturer of personal computer

operating systems could not be held liablefor attempted monopolization of the Inter-net browser market, absent determinationof relevant market; no evidence was identi-fied as to what constituted a browser andwhy other products were not reasonablesubstitutes. Sherman Act, § 2, as amend-ed, 15 U.S.C.A. § 2.

48. Monopolies O12(1.3)A court’s evaluation of an attempted

monopolization claim must include a defini-tion of the relevant market; such a defini-tion establishes a context for evaluatingthe defendant’s actions as well as formeasuring whether the challenged conductpresented a dangerous probability of mo-nopolization. Sherman Act, § 2, asamended, 15 U.S.C.A. § 2.

49. Monopolies O28(8)Determination of a relevant market is

a factual question to be resolved by theDistrict Court in attempted monopolizationaction. Sherman Act, § 2, as amended, 15U.S.C.A. § 2.

50. Monopolies O12(1.3)Because a firm cannot possess monop-

oly power in a market unless that marketis also protected by significant barriers toentry, a firm cannot threaten to achievemonopoly power in a market, for purposes

of attempted monopolization claim, unlessthat market is, or will be, similarly protect-ed. Sherman Act, § 2, as amended, 15U.S.C.A. § 2.


Even assuming that Internet browsermarket was adequately defined, manufac-turer of personal computer operating sys-tem could not be held liable for attemptedmonopolization of browser market, absentevidence that manufacturer would likelyerect significant barriers to entry into thatmarket upon acquisition of a dominantmarket share. Sherman Act, § 2, asamended, 15 U.S.C.A. § 2.

52. Monopolies O12(1.10)Rule of reason, rather than per se

analysis, should govern the legality of ty-ing arrangements involving platform soft-ware products. Sherman Act, § 2, asamended, 15 U.S.C.A. § 2.

53. Monopolies O17.5(2)Rule of reason, rather than per se

analysis, applied to claim that manufactur-er’s contractual and technological bundlingof its Internet browser with its personalcomputer operating system resulted in atying arrangement. Sherman Act, § 2, asamended, 15 U.S.C.A. § 2.

54. Monopolies O17.5(2)There are four elements to a per se

tying violation: (1) the tying and tiedgoods are two separate products; (2) thedefendant has market power in the tyingproduct market; (3) the defendant affordsconsumers no choice but to purchase thetied product from it; and (4) the tyingarrangement forecloses a substantial vol-ume of commerce. Sherman Act, § 2, asamended, 15 U.S.C.A. § 2.

55. Federal Courts O951.1To show on remand that manufactur-

er’s contractual and technological bundling


of its Internet browser with its personalcomputer operating system resulted in atying arrangement under rule of reason,plaintiffs were precluded from arguing anytheory of harm that depended on a precisedefinition of browsers or barriers to entryother than what may have been implicit inthe alleged tying arrangement, whereplaintiffs had failed to provide both a defi-nition of the browser market and barriersto entry to that market as part of theirattempted monopolization claim. ShermanAct, §§ 1, 2, as amended, 15 U.S.C.A.§§ 1, 2.

56. Federal Courts O951.1

To show on remand that manufactur-er’s contractual and technological bundlingof its Internet browser with its personalcomputer operating system resulted in atying arrangement under rule of reason,plaintiffs were required to demonstratethat benefits, if any, of manufacturer’spractices were outweighed by the harms inthe tied product market. Sherman Act,§ 1, as amended, 15 U.S.C.A. § 1.

57. Monopolies O17.5(13)

Manufacturer’s alleged price bundlingof its Internet browser with its personalcomputer operating system could result ina tying arrangement under rule of reasonif it were determined on remand that therewas a positive price increment in operatingsystem associated with browser and thatanticompetitive effects of price bundlingoutweighed any procompetitive justifica-tions. Sherman Act, § 1, as amended, 15U.S.C.A. § 1.

58. Federal Civil Procedure O1992, 2251

Adopting an expedited trial scheduleand receiving evidence through summarywitnesses were with District Court’s dis-cretion in antitrust action against manufac-turer of personal computer operating sys-tem, where case was tried to court.

59. Federal Civil Procedure O2251Trial courts have extraordinarily

broad discretion to determine the mannerin which they will conduct trials; this isparticularly true in a case where the pro-ceedings are being tried to the court with-out a jury.

60. Federal Courts O891Where the proceedings are being tried

to the court without a jury, an appellatecourt will not interfere with the trialcourt’s exercise of its discretion to controlits docket and dispatch its business exceptupon the clearest showing that the proce-dures have resulted in actual and substan-tial prejudice to the complaining litigant.

61. Federal Civil Procedure O1951Factual disputes must be heard in

open court and resolved through trial-likeevidentiary proceedings.

62. Monopolies O28(8)Evidentiary hearing was required

during remedies phase of antitrust actionagainst manufacturer of personal computeroperating system, where parties disputed anumber of facts during remedies phase,including the feasibility of dividing manu-facturer, the likely impact on consumers,and the effect of divestiture on sharehold-ers, and manufacturer repeatedly assertedits right to an evidentiary hearing andsubmitted two offers of proof.

63. Federal Civil Procedure O1951Party has the right to judicial resolu-

tion of disputed facts not just as to theliability phase, but also as to appropriaterelief.

64. Federal Civil Procedure O1951A hearing on the merits—i.e., a trial

on liability—does not substitute for a re-lief-specific evidentiary hearing unless thematter of relief was part of the trial onliability, or unless there are no disputed


factual issues regarding the matter of re-lief.


Remedies decree in antitrust casemust be vacated whenever there is a bonafide disagreement concerning substantiveitems of relief which could be resolved onlyby trial.


Claimed surprise at the districtcourt’s decision to consider permanent in-junctive relief does not, alone, merit rever-sal.

67. Monopolies O28(8)

Remedies decree requiring divestitureof personal computer operating systemmanufacturer found to violate antitrustlaws was not supported by adequate expla-nation, where decree did not discuss rele-vant objectives for such decrees.


Determination on appeal that manu-facturer of personal computer operatingsystem did not commit attempted monopo-lization violation and that remand was re-quired on tying claim required vacation ofremedies decree requiring manufacturer’sdivestiture, where decree was based inpart on findings that manufacturer hadcommitted attempted monopolization andtying violations. Sherman Act, §§ 1, 2, asamended, 15 U.S.C.A. §§ 1, 2.


Where sweeping equitable relief isemployed to remedy multiple antitrust vio-lations, and some of the findings of remedi-able violations do not withstand appellatescrutiny, it is necessary to vacate the rem-edy decree since the implicit findings ofcausal connection no longer exist to war-rant deferential affirmance.

70. Federal Civil Procedure O2582

Generally, a district court is affordedbroad discretion to enter that relief it cal-culates will best remedy the conduct it hasfound to be unlawful.


Motion to disqualify district judgewho presided over antitrust action againstmanufacturer of personal computer operat-ing system could be considered for firsttime on appeal, even though motion wasbased on press accounts of judge’s com-ments about case, rather than on recordevidence, where plaintiffs did not disputecomments attributed to judge in the press,and did not request evidentiary hearing.

72. Judges O49(1)

Federal disqualification provisions re-flect a strong federal policy to preservethe actual and apparent impartiality of thefederal judiciary; judicial misconduct mayimplicate that policy regardless of themeans by which it is disclosed to the pub-lic.


Matter of what questions may be tak-en up and resolved for the first time onappeal is one left primarily to the discre-tion of the courts of appeals, to be exer-cised on the facts of individual cases.

74. Judges O11(2)

District Judge who presided over anti-trust action against manufacturer of per-sonal computer operating system violatedCode of Conduct for United States Judgesby making comments about factual andlegal aspects of case in secret interviewswith reporters while case was pending;comments violated canons forbiddingjudges from commenting publicly on mer-its of a pending or impending action, fromconsidering ex parte communications, andrequiring the avoidance of an appearance


of impropriety. ABA Code of Jud.Con-duct, Canons 2, 3, subd. A(4, 6).

75. Judges O49(2)District judge’s interviews with re-

porters, during which he commented aboutpending antitrust action against manufac-turer of personal computer operating sys-tem, created appearance that he was notacting impartially, within meaning of dis-qualification statute; members of publiccould reasonably question whether judge’sdesire for press coverage influenced hisjudgments. 28 U.S.C.A. § 455(a).

76. Judges O49(2), 56Appearance that district judge was

not acting impartially, created when hecommented about pending antitrust actionagainst manufacturer of personal computeroperating system during secret interviewswith reporters, required judge’s disqualifi-cation retroactive to date he entered reme-dial order, rather than retroactive to anearlier part of the proceedings; full retro-active disqualification would have undulypenalized plaintiffs, who were unaware ofthe misconduct, and manufacturer neitheralleged nor demonstrated that judge’s mis-conduct rose to level of actual bias orprejudice. 28 U.S.C.A. § 455(a).

77. Judges O56At a minimum, statute making dis-

qualification mandatory for conduct thatcalls a judge’s impartiality into questionrequires prospective disqualification of theoffending judge, that is, disqualificationfrom the judge’s hearing any further pro-ceedings in the case. 28 U.S.C.A.§ 455(a).

78. Judges O56There need not be a draconian reme-

dy for every violation of statute makingdisqualification mandatory for conduct thatcalls a judge’s impartiality into question.28 U.S.C.A. § 455(a).

79. Federal Civil Procedure O1969 Federal Courts O856

District judge’s findings of fact in an-titrust action against manufacturer of per-sonal computer operating system weresubject to clearly erroneous review, al-though judge’s comments during inter-views with reporters while case was stillpending created appearance that he wasnot acting impartially. Fed.Rules Civ.Proc.Rule 52(a), 28 U.S.C.A.

80. Federal Courts O776, 850.1There is no de novo appellate review

of fact findings and no intermediate levelbetween de novo and clear error, not evenfor findings the Court of Appeals mayconsider sub-par. Fed.Rules Civ.Proc.Rule 52(a), 28 U.S.C.A.

81. Federal Courts O850.1, 932.1, 941Mandatory nature of rule requiring

review of district court’s fact findings forclear error does not compel Court of Ap-peals to accept fact findings that resultfrom the district court’s misapplication ofgoverning law or that otherwise do notpermit meaningful appellate review, normust Court of Appeals accept findings thatare utterly deficient in other ways; in sucha case, Court of Appeals vacates and re-mands for further factfinding. Fed.RulesCiv.Proc.Rule 52(a), 28 U.S.C.A.

82. Federal Courts O850.1, 932.1When there is fair room for argument

that the district court’s fact findingsshould be vacated in toto, the Court ofAppeals should be especially careful in de-termining that the findings are worthy ofthe deference the clear error standard ofreview prescribes. Fed.Rules Civ.Proc.Rule 52(a), 28 U.S.C.A.

Appeals from the United States DistrictCourt for the District of Columbia (No.98cv01232) (No. 98cv01233).


Richard J. Urowsky and Steven L. Hol-ley argued the causes for appellant. Withthem on the briefs were John L. Warden,Richard C. Pepperman, II, William H.Neukom, Thomas W. Burt, David A. Hein-er, Jr., Charles F. Rule, Robert A. Long,Jr., and Carter G. Phillips. Christopher J.Meyers entered an appearance.

Lars H. Liebeler, Griffin B. Bell, LloydN. Cutler, Louis R. Cohen, C. BoydenGray, William J. Kolasky, William F. Ad-kinson, Jr., Jeffrey D. Ayer, and Jay V.Prabhu were on the brief of amici curiaeThe Association for Competitive Technolo-gy and Computing Technology IndustryAssociation in support of appellant.

David R. Burton was on the brief foramicus curiae Center for the Moral De-fense of Capitalism in support of appellant.

Robert S. Getman was on the brief foramicus curiae Association for ObjectiveLaw in support of appellant.

Jeffrey P. Minear and David C. Freder-ick, Assistants to the Solicitor General,United States Department of Justice, andJohn G. Roberts, Jr., argued the causes forappellees. With them on the brief were A.Douglas Melamed, Acting Assistant Attor-ney General, United States Department ofJustice, Jeffrey H. Blattner, Deputy Assis-tant Attorney General, Mary Jean Molten-

brey, Director, Catherine G. O’Sullivan,Robert B. Nicholson, Adam D. Hirsh, An-drea Limmer, David Seidman, and Chris-topher Sprigman, Attorneys, Eliot Spitzer,Attorney General, State of New York,Richard L. Schwartz, Assistant AttorneyGeneral, and Kevin J. O’Connor, Office ofthe Attorney General, State of Wisconsin.

John Rogovin, Kenneth W. Starr, JohnF. Wood, Elizabeth Petrela, Robert H.Bork, Theodore W. Ullyot, Jason M. Mah-ler, Stephen M. Shapiro, Donald M. Falk,Mitchell S. Pettit, Kevin J. Arquit, andMichael C. Naughton were on the brief foramici curiae America Online, Inc., et al., insupport of appellee. Paul T. Cappuccioentered an appearance.

Lee A. Hollaar, appearing pro se, wason the brief for amicus curiae Lee A.Hollaar.

Carl Lundgren, appearing pro se, wason the brief for amicus curiae Carl Lund-gren.

Before: EDWARDS, Chief Judge,WILLIAMS, GINSBURG, SENTELLE,RANDOLPH, ROGERS and TATEL,Circuit Judges.

Opinion for the Court filed PERCURIAM.

TABLE OF CONTENTS

SummaryTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTT44

I. Introduction TTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTT47A. Background TTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTT47B. Overview TTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTT48

II. MonopolizationTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTT50A. Monopoly PowerTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTT51

1. Market Structure TTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTT51a. Market definition TTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTT51b. Market power TTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTT54

2. Direct Proof TTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTT56B. Anticompetitive ConductTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTT58

1. Licenses Issued to Original Equipment Manufacturers TTTTTTTTTTTTTTTTTTT59a. Anticompetitive effect of the license restrictions TTTTTTTTTTTTTTTTTTTTT60b. Microsoft’s justifications for the license restrictions TTTTTTTTTTTTTTTTTT62

2. Integration of IE and Windows TTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTT64a. Anticompetitive effect of integrationTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTT65


b. Microsoft’s justifications for integrationTTTTTTTTTTTTTTTTTTTTTTTTTTTTT663. Agreements with Internet Access Providers TTTTTTTTTTTTTTTTTTTTTTTTTTTT674. Dealings with Internet Content Providers, Independent Software

Vendors, and Apple Computer TTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTT715. Java TTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTT74

a. The incompatible JVM TTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTT74b. The First Wave Agreements TTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTT75c. Deception of Java developersTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTT76d. The threat to Intel TTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTT77

6. Course of ConductTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTT78C. CausationTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTT78

III. Attempted Monopolization TTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTT80A. Relevant MarketTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTT81B. Barriers to Entry TTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTT82

IV. Tying TTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTT84A. Separate–Products Inquiry Under the Per Se Test TTTTTTTTTTTTTTTTTTTTTTTTTT85B. Per Se Analysis Inappropriate for this Case TTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTT89C. On Remand TTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTT95

V. Trial Proceedings and Remedy TTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTT97A. Factual BackgroundTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTT98B. Trial Proceedings TTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTT100C. Failure to Hold an Evidentiary Hearing TTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTT101D. Failure to Provide an Adequate ExplanationTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTT103E. Modification of Liability TTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTT103F. On Remand TTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTT105G. Conclusion TTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTT107

VI. Judicial MisconductTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTT107A. The District Judge’s Communications with the Press TTTTTTTTTTTTTTTTTTTTTTT107B. Violations of the Code of Conduct for United States Judges TTTTTTTTTTTTTTTTT111C. Appearance of Partiality TTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTT114D. Remedies for Judicial Misconduct and Appearance of Partiality TTTTTTTTTTTTTT116

1. Disqualification TTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTT1162. Review of Findings of Fact and Conclusions of LawTTTTTTTTTTTTTTTTTTTTT117

VII. Conclusion TTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTT118

PER CURIAM:

Microsoft Corporation appeals fromjudgments of the District Court findingthe company in violation of §§ 1 and 2 ofthe Sherman Act and ordering variousremedies.

The action against Microsoft arose pur-suant to a complaint filed by the UnitedStates and separate complaints filed byindividual States. The District Court de-termined that Microsoft had maintained amonopoly in the market for Intelcompati-ble PC operating systems in violation of§ 2; attempted to gain a monopoly in themarket for internet browsers in violationof § 2; and illegally tied two purportedlyseparate products, Windows and InternetExplorer (‘‘IE’’), in violation of § 1. Unit-

ed States v. Microsoft Corp., 87 F.Supp.2d30 (D.D.C.2000) (‘‘Conclusions of Law’’).The District Court then found that thesame facts that established liability under§§ 1 and 2 of the Sherman Act mandatedfindings of liability under analogous statelaw antitrust provisions. Id. To remedythe Sherman Act violations, the DistrictCourt issued a Final Judgment requiringMicrosoft to submit a proposed plan ofdivestiture, with the company to be splitinto an operating systems business and anapplications business. United States v.Microsoft Corp., 97 F.Supp.2d 59, 64–65(D.D.C.2000) (‘‘Final Judgment’’). TheDistrict Court’s remedial order also con-tains a number of interim restrictions onMicrosoft’s conduct. Id. at 66–69.


Microsoft’s appeal contests both the le-gal conclusions and the resulting remedialorder. There are three principal aspectsof this appeal. First, Microsoft challengesthe District Court’s legal conclusions as toall three alleged antitrust violations andalso a number of the procedural and factu-al foundations on which they rest. Second,Microsoft argues that the remedial ordermust be set aside, because the DistrictCourt failed to afford the company an evi-dentiary hearing on disputed facts and,also, because the substantive provisions ofthe order are flawed. Finally, Microsoftasserts that the trial judge committed ethi-cal violations by engaging in impermissibleex parte contacts and making inappropri-ate public comments on the merits of thecase while it was pending. Microsoft ar-gues that these ethical violations compro-mised the District Judge’s appearance ofimpartiality, thereby necessitating his dis-qualification and vacatur of his Findings ofFact, Conclusions of Law, and Final Judg-ment.

After carefully considering the volumi-nous record on appeal—including the Dis-trict Court’s Findings of Fact and Conclu-sions of Law, the testimony and exhibitssubmitted at trial, the parties’ briefs, andthe oral arguments before this court—wefind that some but not all of Microsoft’sliability challenges have merit. According-ly, we affirm in part and reverse in partthe District Court’s judgment that Micro-soft violated § 2 of the Sherman Act byemploying anticompetitive means to main-tain a monopoly in the operating systemmarket; we reverse the District Court’sdetermination that Microsoft violated § 2of the Sherman Act by illegally attemptingto monopolize the internet browser mar-ket; and we remand the District Court’sfinding that Microsoft violated § 1 of theSherman Act by unlawfully tying its brow-ser to its operating system. Our judgmentextends to the District Court’s findings

with respect to the state law counterpartsof the plaintiffs’ Sherman Act claims.

We also find merit in Microsoft’s chal-lenge to the Final Judgment embracingthe District Court’s remedial order.There are several reasons supporting thisconclusion. First, the District Court’s Fi-nal Judgment rests on a number of liabil-ity determinations that do not survive ap-pellate review; therefore, the remedialorder as currently fashioned cannot stand.Furthermore, we would vacate and re-mand the remedial order even were we touphold the District Court’s liability deter-minations in their entirety, because theDistrict Court failed to hold an evidentia-ry hearing to address remedies-specificfactual disputes.

Finally, we vacate the Final Judgmenton remedies, because the trial judge en-gaged in impermissible ex parte contactsby holding secret interviews with membersof the media and made numerous offensivecomments about Microsoft officials in pub-lic statements outside of the courtroom,giving rise to an appearance of partiality.Although we find no evidence of actualbias, we hold that the actions of the trialjudge seriously tainted the proceedings be-fore the District Court and called intoquestion the integrity of the judicial pro-cess. We are therefore constrained to va-cate the Final Judgment on remedies, re-mand the case for reconsideration of theremedial order, and require that the casebe assigned to a different trial judge onremand. We believe that this dispositionwill be adequate to cure the cited impro-prieties.

In sum, for reasons more fully explainedbelow, we affirm in part, reverse in part,and remand in part the District Court’sjudgment assessing liability. We vacate infull the Final Judgment embodying theremedial order and remand the case to a


different trial judge for further proceed-ings consistent with this opinion.

I. INTRODUCTION

A. Background

In July 1994, officials at the Departmentof Justice (‘‘DOJ’’), on behalf of the UnitedStates, filed suit against Microsoft, charg-ing the company with, among other things,unlawfully maintaining a monopoly in theoperating system market through anticom-petitive terms in its licensing and softwaredeveloper agreements. The parties subse-quently entered into a consent decree, thusavoiding a trial on the merits. See UnitedStates v. Microsoft Corp., 56 F.3d 1448(D.C.Cir.1995) (‘‘Microsoft I’’). Threeyears later, the Justice Department filed acivil contempt action against Microsoft forallegedly violating one of the decree’s pro-visions. On appeal from a grant of apreliminary injunction, this court held thatMicrosoft’s technological bundling of IE3.0 and 4.0 with Windows 95 did not violatethe relevant provision of the consent de-cree. United States v. Microsoft Corp.,147 F.3d 935 (D.C.Cir.1998) (‘‘MicrosoftII’’). We expressly reserved the questionwhether such bundling might independent-ly violate §§ 1 or 2 of the Sherman Act.Id. at 950 n. 14.

On May 18, 1998, shortly before issuanceof the Microsoft II decision, the UnitedStates and a group of State plaintiffs filedseparate (and soon thereafter consolidat-ed) complaints, asserting antitrust viola-tions by Microsoft and seeking preliminaryand permanent injunctions against thecompany’s allegedly unlawful conduct.The complaints also sought any ‘‘other pre-liminary and permanent relief as is neces-sary and appropriate to restore competi-tive conditions in the markets affected byMicrosoft’s unlawful conduct.’’ Gov’t’sCompl. at 53, United States v. MicrosoftCorp., No. 98–1232 (D.D.C.1999). Relying

almost exclusively on Microsoft’s variedefforts to unseat Netscape Navigator asthe preeminent internet browser, plaintiffscharged four distinct violations of theSherman Act: (1) unlawful exclusive deal-ing arrangements in violation of § 1; (2)unlawful tying of IE to Windows 95 andWindows 98 in violation of § 1; (3) unlaw-ful maintenance of a monopoly in the PCoperating system market in violation of§ 2; and (4) unlawful attempted monopoli-zation of the internet browser market inviolation of § 2. The States also broughtpendent claims charging Microsoft with vi-olations of various State antitrust laws.

The District Court scheduled the caseon a ‘‘fast track.’’ The hearing on the pre-liminary injunction and the trial on themerits were consolidated pursuant to FED.

R.CIV.P. 65(a)(2). The trial was thenscheduled to commence on September 8,1998, less than four months after the com-plaints had been filed. In a series ofpretrial orders, the District Court limitedeach side to a maximum of 12 trial wit-nesses plus two rebuttal witnesses. It re-quired that all trial witnesses’ direct testi-mony be submitted to the court in theform of written declarations. The DistrictCourt also made allowances for the use ofdeposition testimony at trial to prove sub-ordinate or predicate issues. Followingthe grant of three brief continuances, thetrial started on October 19, 1998.

After a 76–day bench trial, the DistrictCourt issued its Findings of Fact. UnitedStates v. Microsoft Corp., 84 F.Supp.2d 9(D.D.C.1999) (‘‘Findings of Fact’’). Thistriggered two independent courses of ac-tion. First, the District Court establisheda schedule for briefing on possible legalconclusions, inviting Professor LawrenceLessig to participate as amicus curiae.Second, the District Court referred thecase to mediation to afford the parties anopportunity to settle their differences. The


Honorable Richard A. Posner, Chief Judgeof the United States Court of Appeals forthe Seventh Circuit, was appointed toserve as mediator. The parties concurredin the referral to mediation and in thechoice of mediator.

Mediation failed after nearly fourmonths of settlement talks between theparties. On April 3, 2000, with the parties’briefs having been submitted and consid-ered, the District Court issued its conclu-sions of law. The District Court foundMicrosoft liable on the § 1 tying and § 2monopoly maintenance and attempted mo-nopolization claims, Conclusions of Law, at35–51, while ruling that there was insuffi-cient evidence to support a § 1 exclusivedealing violation, id. at 51–54. As to thependent State actions, the District Courtfound the State antitrust laws contermi-nous with §§ 1 and 2 of the Sherman Act,thereby obviating the need for furtherState-specific analysis. Id. at 54–56. Inthose few cases where a State’s law re-quired an additional showing of intrastateimpact on competition, the District Courtfound the requirement easily satisfied onthe evidence at hand. Id. at 55.

Having found Microsoft liable on all butone count, the District Court then askedplaintiffs to submit a proposed remedy.Plaintiffs’ proposal for a remedial orderwas subsequently filed within four weeks,along with six supplemental declarationsand over 50 new exhibits. In their propos-al, plaintiffs sought specific conduct reme-dies, plus structural relief that would splitMicrosoft into an applications companyand an operating systems company. TheDistrict Court rejected Microsoft’s requestfor further evidentiary proceedings and,following a single hearing on the merits ofthe remedy question, issued its FinalJudgment on June 7, 2000. The DistrictCourt adopted plaintiffs’ proposed remedywithout substantive change.

Microsoft filed a notice of appeal withina week after the District Court issued itsFinal Judgment. This court then orderedthat any proceedings before it be heard bythe court sitting en banc. Before anysubstantive matters were addressed bythis court, however, the District Court cer-tified appeal of the case brought by theUnited States directly to the SupremeCourt pursuant to 15 U.S.C. § 29(b), whilestaying the final judgment order in thefederal and state cases pending appeal.The States thereafter petitioned the Su-preme Court for a writ of certiorari intheir case. The Supreme Court declinedto hear the appeal of the Government’scase and remanded the matter to thiscourt; the Court likewise denied theStates’ petition for writ of certiorari. Mi-crosoft Corp. v. United States, 530 U.S.1301, 121 S.Ct. 25, 147 L.Ed.2d 1048(2000). This consolidated appeal followed.

B. Overview

Before turning to the merits of Micro-soft’s various arguments, we pause to re-flect briefly on two matters of note, onepractical and one theoretical.

The practical matter relates to the tem-poral dimension of this case. The litiga-tion timeline in this case is hardly proble-matic. Indeed, it is noteworthy that a caseof this magnitude and complexity has pro-ceeded from the filing of complaintsthrough trial to appellate decision in amere three years. See, e.g., Data Gen.Corp. v. Grumman Sys. Support Corp., 36F.3d 1147, 1155 (1st Cir.1994) (six yearsfrom filing of complaint to appellate deci-sion); Transamerica Computer Co., Inc. v.IBM, 698 F.2d 1377, 1381 (9th Cir.1983)(over four years from start of trial toappellate decision); United States v. Unit-ed Shoe Mach. Corp., 110 F.Supp. 295, 298(D.Mass.1953) (over five years from filingof complaint to trial court decision).


What is somewhat problematic, however,is that just over six years have passedsince Microsoft engaged in the first con-duct plaintiffs allege to be anticompetitive.As the record in this case indicates, sixyears seems like an eternity in the com-puter industry. By the time a court canassess liability, firms, products, and themarketplace are likely to have changeddramatically. This, in turn, threatensenormous practical difficulties for courtsconsidering the appropriate measure of re-lief in equitable enforcement actions, bothin crafting injunctive remedies in the firstinstance and reviewing those remedies inthe second. Conduct remedies may beunavailing in such cases, because innova-tion to a large degree has already ren-dered the anticompetitive conduct obsolete(although by no means harmless). Andbroader structural remedies present theirown set of problems, including how a courtgoes about restoring competition to a dra-matically changed, and constantly chang-ing, marketplace. That is just one reasonwhy we find the District Court’s refusal inthe present case to hold an evidentiaryhearing on remedies—to update and fleshout the available information before seri-ously entertaining the possibility of dra-matic structural relief—so problematic.See infra Section V.

We do not mean to say that enforcementactions will no longer play an importantrole in curbing infringements of the anti-trust laws in technologically dynamic mar-kets, nor do we assume this in assessingthe merits of this case. Even in thosecases where forward-looking remedies ap-pear limited, the Government will continueto have an interest in defining the contoursof the antitrust laws so that law-abidingfirms will have a clear sense of what ispermissible and what is not. And thethreat of private damage actions will re-main to deter those firms inclined to testthe limits of the law.

The second matter of note is more theo-retical in nature. We decide this caseagainst a backdrop of significant debateamongst academics and practitioners overthe extent to which ‘‘old economy’’ § 2monopolization doctrines should apply tofirms competing in dynamic technologicalmarkets characterized by network effects.In markets characterized by network ef-fects, one product or standard tends to-wards dominance, because ‘‘the utility thata user derives from consumption of thegood increases with the number of otheragents consuming the good.’’ Michael L.Katz & Carl Shapiro, Network Externali-ties, Competition, and Compatibility, 75AM. ECON. REV. 424, 424 (1985). For exam-ple, ‘‘[a]n individual consumer’s demand touse (and hence her benefit from) the tele-phone network TTT increases with thenumber of other users on the networkwhom she can call or from whom she canreceive calls.’’ Howard A. Shelanski & J.Gregory Sidak, Antitrust Divestiture inNetwork Industries, 68 U. CHI. L. REV. 1, 8(2001). Once a product or standardachieves wide acceptance, it becomes moreor less entrenched. Competition in suchindustries is ‘‘for the field’’ rather than‘‘within the field.’’ See Harold Demsetz,Why Regulate Utilities?, 11 J.L. & ECON.55, 57 & n.7 (1968) (emphasis omitted).

In technologically dynamic markets,however, such entrenchment may be tem-porary, because innovation may alter thefield altogether. See JOSEPH A. SCHUMPET-

ER, CAPITALISM, SOCIALISM AND DEMOCRACY

81–90 (Harper Perennial 1976) (1942).Rapid technological change leads to mar-kets in which ‘‘firms compete through in-novation for temporary market dominance,from which they may be displaced by thenext wave of product advancements.’’Shelanski & Sidak, at 11–12 (discussingSchumpeterian competition, which pro-ceeds ‘‘sequentially over time rather than


simultaneously across a market’’). Micro-soft argues that the operating system mar-ket is just such a market.

Whether or not Microsoft’s characteriza-tion of the operating system market iscorrect does not appreciably alter our mis-sion in assessing the alleged antitrust vio-lations in the present case. As an initialmatter, we note that there is no consensusamong commentators on the question ofwhether, and to what extent, current mo-nopolization doctrine should be amended toaccount for competition in technologicallydynamic markets characterized by net-work effects. Compare Steven C. Salop &R. Craig Romaine, Preserving Monopoly:Economic Analysis, Legal Standards, andMicrosoft, 7 GEO. MASON L. REV. 617, 654–55, 663–64 (1999) (arguing that exclusion-ary conduct in high-tech networked indus-tries deserves heightened antitrust scruti-ny in part because it may threaten to deterinnovation), with Ronald A. Cass & KeithN. Hylton, Preserving Competition: Eco-nomic Analysis, Legal Standards and Mi-crosoft, 8 GEO. MASON L. REV. 1, 36–39(1999) (equivocating on the antitrust impli-cations of network effects and noting thatthe presence of network externalities mayactually encourage innovation by guaran-teeing more durable monopolies to inno-vating winners). Indeed, there is somesuggestion that the economic consequencesof network effects and technological dy-namism act to offset one another, therebymaking it difficult to formulate categoricalantitrust rules absent a particularizedanalysis of a given market. See Shelanski& Sidak, at 6–7 (‘‘High profit marginsmight appear to be the benign and neces-sary recovery of legitimate investment re-turns in a Schumpeterian framework, butthey might represent exploitation of cus-tomer lock-in and monopoly power whenviewed through the lens of network eco-nomicsTTTT The issue is particularly com-plex because, in network industries charac-

terized by rapid innovation, both forcesmay be operating and can be difficult toisolate.’’).

Moreover, it should be clear that Micro-soft makes no claim that anticompetitiveconduct should be assessed differently intechnologically dynamic markets. Itclaims only that the measure of monopolypower should be different. For reasonsfully discussed below, we reject Microsoft’smonopoly power argument. See infra Sec-tion II.A.

With this backdrop in mind, we turn tothe specific challenges raised in Microsoft’sappeal.

II. MONOPOLIZATION

[1, 2] Section 2 of the Sherman Actmakes it unlawful for a firm to ‘‘monopo-lize.’’ 15 U.S.C. § 2. The offense of mo-nopolization has two elements: ‘‘(1) thepossession of monopoly power in the rele-vant market and (2) the willful acquisitionor maintenance of that power as distin-guished from growth or development as aconsequence of a superior product, busi-ness acumen, or historic accident.’’ Unit-ed States v. Grinnell Corp., 384 U.S. 563,570–71, 86 S.Ct. 1698, 16 L.Ed.2d 778(1966). The District Court applied thistest and found that Microsoft possessesmonopoly power in the market for Intel-compatible PC operating systems. Focus-ing primarily on Microsoft’s efforts to sup-press Netscape Navigator’s threat to itsoperating system monopoly, the court alsofound that Microsoft maintained its powernot through competition on the merits, butthrough unlawful means. Microsoft chal-lenges both conclusions. We defer to theDistrict Court’s findings of fact, settingthem aside only if clearly erroneous. FED

R. CIV. P. 52(a). We review legal questionsde novo. United States ex rel. Modern


Elec., Inc. v. Ideal Elec. Sec. Co., 81 F.3d240, 244 (D.C.Cir.1996).

We begin by considering whether Micro-soft possesses monopoly power, see infraSection II.A, and finding that it does, weturn to the question whether it maintainedthis power through anticompetitive means.Agreeing with the District Court that thecompany behaved anticompetitively, seeinfra Section II.B, and that these actionscontributed to the maintenance of its mo-nopoly power, see infra Section II.C, weaffirm the court’s finding of liability formonopolization.

A. Monopoly Power

[3–5] While merely possessing monop-oly power is not itself an antitrust viola-tion, see Northeastern Tel. Co. v. AT & T,651 F.2d 76, 84–85 (2d Cir.1981), it is anecessary element of a monopolizationcharge, see Grinnell, 384 U.S. at 570, 86S.Ct. 1698. The Supreme Court definesmonopoly power as ‘‘the power to controlprices or exclude competition.’’ UnitedStates v. E.I. du Pont de Nemours & Co.,351 U.S. 377, 391, 76 S.Ct. 994, 100 L.Ed.1264 (1956). More precisely, a firm is amonopolist if it can profitably raise pricessubstantially above the competitive level.2A PHILLIP E. AREEDA ET AL., ANTITRUST LAW

¶ 501, at 85 (1995); cf. Ball Mem’l Hosp.,Inc. v. Mut. Hosp. Ins., Inc., 784 F.2d1325, 1335 (7th Cir.1986) (defining marketpower as ‘‘the ability to cut back the mar-ket’s total output and so raise price’’).Where evidence indicates that a firm hasin fact profitably done so, the existence ofmonopoly power is clear. See Rebel OilCo. v. Atl. Richfield Co., 51 F.3d 1421, 1434(9th Cir.1995); see also FTC v. IndianaFed’n of Dentists, 476 U.S. 447, 460–61,106 S.Ct. 2009, 90 L.Ed.2d 445 (1986) (us-ing direct proof to show market power inSherman Act § 1 unreasonable restraint oftrade action). Because such direct proof is

only rarely available, courts more typicallyexamine market structure in search of cir-cumstantial evidence of monopoly power.2A AREEDA ET AL., ANTITRUST LAW ¶ 531a, at156; see also, e.g., Grinnell, 384 U.S. at571, 86 S.Ct. 1698. Under this structuralapproach, monopoly power may be in-ferred from a firm’s possession of a domi-nant share of a relevant market that isprotected by entry barriers. See RebelOil, 51 F.3d at 1434. ‘‘Entry barriers’’ arefactors (such as certain regulatory require-ments) that prevent new rivals from timelyresponding to an increase in price abovethe competitive level. See S. Pac. Com-munications Co. v.AT & T, 740 F.2d 980,1001–02 (D.C.Cir.1984).

The District Court considered thesestructural factors and concluded that Mi-crosoft possesses monopoly power in a rel-evant market. Defining the market asIntel-compatible PC operating systems,the District Court found that Microsoft hasa greater than 95% share. It also foundthe company’s market position protectedby a substantial entry barrier. Conclu-sions of Law, at 36.

Microsoft argues that the District Courtincorrectly defined the relevant market.It also claims that there is no barrier toentry in that market. Alternatively, Mi-crosoft argues that because the softwareindustry is uniquely dynamic, direct proof,rather than circumstantial evidence, moreappropriately indicates whether it possess-es monopoly power. Rejecting each argu-ment, we uphold the District Court’s find-ing of monopoly power in its entirety.

1. Market Structure

a. Market definition

[6, 7] ‘‘Because the ability of consum-ers to turn to other suppliers restrains afirm from raising prices above the compet-itive level,’’ Rothery Storage & Van Co. v.Atlas Van Lines, Inc., 792 F.2d 210, 218


(D.C.Cir.1986), the relevant market mustinclude all products ‘‘reasonably inter-changeable by consumers for the samepurposes.’’ du Pont, 351 U.S. at 395, 76S.Ct. 994. In this case, the District Courtdefined the market as ‘‘the licensing of allIntel-compatible PC operating systemsworldwide,’’ finding that there are ‘‘cur-rently no products—and TTT there are notlikely to be any in the near future—that asignificant percentage of computer usersworldwide could substitute for [these oper-ating systems] without incurring substan-tial costs.’’ Conclusions of Law, at 36.Calling this market definition ‘‘far too nar-row,’’ Appellant’s Opening Br. at 84, Mi-crosoft argues that the District Court im-properly excluded three types of products:non-Intel compatible operating systems(primarily Apple’s Macintosh operatingsystem, Mac OS), operating systems fornon-PC devices (such as handheld comput-ers and portal websites), and ‘‘middleware’’products, which are not operating systemsat all.

We begin with Mac OS. Microsoft’sargument that Mac OS should have beenincluded in the relevant market suffersfrom a flaw that infects many of the com-pany’s monopoly power claims: the compa-ny fails to challenge the District Court’sfactual findings, or to argue that thesefindings do not support the court’s conclu-sions. The District Court found that con-sumers would not switch from Windows toMac OS in response to a substantial priceincrease because of the costs of acquiringthe new hardware needed to run Mac OS(an Apple computer and peripherals) andcompatible software applications, as wellas because of the effort involved in learn-ing the new system and transferring filesto its format. Findings of Fact ¶ 20. Thecourt also found the Apple system lessappealing to consumers because it costsconsiderably more and supports fewer ap-plications. Id. ¶ 21. Microsoft responds

only by saying: ‘‘the district court’s mar-ket definition is so narrow that it excludesApple’s Mac OS, which has competed withWindows for years, simply because theMac OS runs on a different microproces-sor.’’ Appellant’s Opening Br. at 84. Thisgeneral, conclusory statement falls farshort of what is required to challenge find-ings as clearly erroneous. Pendleton v.Rumsfeld, 628 F.2d 102, 106 (D.C.Cir.1980); see also Terry v. Reno, 101 F.3d1412, 1415 (D.C.Cir.1996) (holding thatclaims made but not argued in a brief arewaived). Microsoft neither points to evi-dence contradicting the District Court’sfindings nor alleges that supporting recordevidence is insufficient. And since Micro-soft does not argue that even if we acceptthese findings, they do not support theDistrict Court’s conclusion, we have nobasis for upsetting the court’s decision toexclude Mac OS from the relevant market.

Microsoft’s challenge to the DistrictCourt’s exclusion of non-PC based compet-itors, such as information appliances (han-dheld devices, etc.) and portal websitesthat host serverbased software applica-tions, suffers from the same defect: thecompany fails to challenge the DistrictCourt’s key factual findings. In particular,the District Court found that because in-formation appliances fall far short of per-forming all of the functions of a PC, mostconsumers will buy them only as a supple-ment to their PCs. Findings of Fact ¶ 23.The District Court also found that portalwebsites do not presently host enough ap-plications to induce consumers to switch,nor are they likely to do so in the nearfuture. Id. ¶ 27. Again, because Micro-soft does not argue that the DistrictCourt’s findings do not support its conclu-sion that information appliances and portalwebsites are outside the relevant market,we adhere to that conclusion.


[8] This brings us to Microsoft’s mainchallenge to the District Court’s marketdefinition: the exclusion of middleware.Because of the importance of middlewareto this case, we pause to explain what it isand how it relates to the issue before us.

Operating systems perform many func-tions, including allocating computer memo-ry and controlling peripherals such asprinters and keyboards. See Direct Testi-mony of Frederick Warren–Boulton ¶ 20,reprinted in 5 J.A. at 3172–73. Operatingsystems also function as platforms for soft-ware applications. They do this by ‘‘ex-posing’’—i.e., making available to softwaredevelopers—routines or protocols that per-form certain widely-used functions. Theseare known as Application ProgrammingInterfaces, or ‘‘APIs.’’ See Direct Testi-mony of James Barksdale ¶ 70, reprintedin 5 J.A. at 2895–96. For example, Win-dows contains an API that enables users todraw a box on the screen. See DirectTestimony of Michael T. Devlin ¶ 12, re-printed in 5 J.A. at 3525. Software devel-opers wishing to include that function in anapplication need not duplicate it in theirown code. Instead, they can ‘‘call’’—i.e.,use—the Windows API. See Direct Testi-mony of James Barksdale ¶ ¶ 70–71, re-printed in 5 J.A. at 2895–97. Windowscontains thousands of APIs, controlling ev-erything from data storage to font display.See Direct Testimony of Michael Devlin¶ 12, reprinted in 5 J.A. at 3525.

Every operating system has differentAPIs. Accordingly, a developer whowrites an application for one operatingsystem and wishes to sell the application tousers of another must modify, or ‘‘port,’’the application to the second operatingsystem. Findings of Fact ¶ 4. This pro-cess is both timeconsuming and expensive.Id. ¶ 30.

‘‘Middleware’’ refers to software prod-ucts that expose their own APIs. Id. ¶ 28;

Direct Testimony of Paul Maritz ¶ ¶ 234–36, reprinted in 6 J.A. at 3727–29. Be-cause of this, a middleware product writ-ten for Windows could take over some orall of Windows’s valuable platform func-tions—that is, developers might begin torely upon APIs exposed by the middlewarefor basic routines rather than relying uponthe API set included in Windows. If mid-dleware were written for multiple operat-ing systems, its impact could be evengreater. The more developers could relyupon APIs exposed by such middleware,the less expensive porting to different op-erating systems would be. Ultimately, ifdevelopers could write applications relyingexclusively on APIs exposed by middle-ware, their applications would run on anyoperating system on which the middlewarewas also present. See Direct Testimony ofAvadis Tevanian, Jr. ¶ 45, reprinted in 5J.A. at 3113. Netscape Navigator andJava—both at issue in this case—are mid-dleware products written for multiple op-erating systems. Findings of Fact ¶ 28.

Microsoft argues that, because middle-ware could usurp the operating system’splatform function and might eventuallytake over other operating system functions(for instance, by controlling peripherals),the District Court erred in excluding Navi-gator and Java from the relevant market.The District Court found, however, thatneither Navigator, Java, nor any othermiddleware product could now, or wouldsoon, expose enough APIs to serve as aplatform for popular applications, muchless take over all operating system func-tions. Id. ¶ ¶ 28–29. Again, Microsoftfails to challenge these findings, insteadsimply asserting middleware’s ‘‘potential’’as a competitor. Appellant’s Opening Br.at 86. The test of reasonable interchange-ability, however, required the DistrictCourt to consider only substitutes thatconstrain pricing in the reasonably fore-


seeable future, and only products that canenter the market in a relatively short timecan perform this function. See Rothery,792 F.2d at 218 (‘‘Because the ability ofconsumers to turn to other suppliers re-strains a firm from raising prices abovethe competitive level, the definition of the‘relevant market’ rests on a determinationof available substitutes.’’); see also Find-ings of Fact ¶ 29 (‘‘[I]t would take severalyears for middleware TTT to evolve’’ into aproduct that can constrain operating sys-tem pricing.). Whatever middleware’s ul-timate potential, the District Court foundthat consumers could not now abandontheir operating systems and switch to mid-dleware in response to a sustained pricefor Windows above the competitive level.Findings of Fact ¶ ¶ 28, 29. Nor is mid-dleware likely to overtake the operatingsystem as the primary platform for soft-ware development any time in the nearfuture. Id.

Alternatively, Microsoft argues that theDistrict Court should not have excludedmiddleware from the relevant market be-cause the primary focus of the plaintiffs’§ 2 charge is on Microsoft’s attempts tosuppress middleware’s threat to its operat-ing system monopoly. According to Mi-crosoft, it is ‘‘contradict[ory],’’ 2/26/2001Ct. Appeals Tr. at 20, to define the rele-vant market to exclude the ‘‘very competi-tive threats that gave rise’’ to the action.Appellant’s Opening Br. at 84. The pur-ported contradiction lies between plaintiffs’§ 2 theory, under which Microsoft pre-served its monopoly against middlewaretechnologies that threatened to become vi-able substitutes for Windows, and its theo-ry of the relevant market, under whichmiddleware is not presently a viable sub-stitute for Windows. Because middle-ware’s threat is only nascent, however, nocontradiction exists. Nothing in § 2 of theSherman Act limits its prohibition to ac-tions taken against threats that are al-

ready well-developed enough to serve aspresent substitutes. See infra SectionII.C. Because market definition is meantto identify products ‘‘reasonably inter-changeable by consumers,’’ du Pont, 351U.S. at 395, 76 S.Ct. 994, and becausemiddleware is not now interchangeablewith Windows, the District Court had goodreason for excluding middleware from therelevant market.

b. Market power

[9] Having thus properly defined therelevant market, the District Court foundthat Windows accounts for a greater than95% share. Findings of Fact ¶ 35. Thecourt also found that even if Mac OS wereincluded, Microsoft’s share would exceed80%. Id. Microsoft challenges neitherfinding, nor does it argue that such amarket share is not predominant. Cf.Grinnell, 384 U.S. at 571, 86 S.Ct. 1698(87% is predominant); Eastman KodakCo. v. Image Technical Servs., Inc., 504U.S. 451, 481, 112 S.Ct. 2072, 119 L.Ed.2d265 (1992) (80%); du Pont, 351 U.S. at 379,391, 76 S.Ct. 994 (75%).

[10] Instead, Microsoft claims thateven a predominant market share does notby itself indicate monopoly power. Al-though the ‘‘existence of [monopoly] powerordinarily may be inferred from the pre-dominant share of the market,’’ Grinnell,384 U.S. at 571, 86 S.Ct. 1698, we agreewith Microsoft that because of the possibil-ity of competition from new entrants, seeBall Mem’l Hosp., Inc., 784 F.2d at 1336,looking to current market share alone canbe ‘‘misleading.’’ Hunt-Wesson Foods,Inc. v. Ragu Foods, Inc., 627 F.2d 919, 924(9th Cir.1980); see also Ball Mem’l Hosp.,Inc., 784 F.2d at 1336 (‘‘Market share re-flects current sales, but today’s sales donot always indicate power over sales andprice tomorrow.’’) In this case, however,the District Court was not misled. Con-


sidering the possibility of new rivals, thecourt focused not only on Microsoft’s pres-ent market share, but also on the structur-al barrier that protects the company’s fu-ture position. Conclusions of Law, at 36.That barrier—the ‘‘applications barrier toentry’’—stems from two characteristics ofthe software market: (1) most consumersprefer operating systems for which a largenumber of applications have already beenwritten; and (2) most developers prefer towrite for operating systems that alreadyhave a substantial consumer base. SeeFindings of Fact ¶ ¶ 30, 36. This ‘‘chick-en-and-egg’’ situation ensures that applica-tions will continue to be written for thealready dominant Windows, which in turnensures that consumers will continue toprefer it over other operating systems.Id.

Challenging the existence of the applica-tions barrier to entry, Microsoft observesthat software developers do write applica-tions for other operating systems, pointingout that at its peak IBM’s OS/2 supportedapproximately 2,500 applications. Id. ¶ 46.This misses the point. That some develop-ers write applications for other operatingsystems is not at all inconsistent with thefinding that the applications barrier to en-try discourages many from writing forthese less popular platforms. Indeed, theDistrict Court found that IBM’s difficultyin attracting a larger number of softwaredevelopers to write for its platform seri-ously impeded OS/2’s success. Id. ¶ 46.

Microsoft does not dispute that Windowssupports many more applications than anyother operating system. It argues insteadthat ‘‘[i]t defies common sense’’ to suggestthat an operating system must support asmany applications as Windows does (morethan 70,000, according to the DistrictCourt, id. ¶ 40) to be competitive. Appel-lant’s Opening Br. at 96. Consumers, Mi-crosoft points out, can only use a very

small percentage of these applications. Id.As the District Court explained, however,the applications barrier to entry gives con-sumers reason to prefer the dominant op-erating system even if they have no needto use all applications written for it:

The consumer wants an operating sys-tem that runs not only types of applica-tions that he knows he will want to use,but also those types in which he mightdevelop an interest later. Also, the con-sumer knows that if he chooses an oper-ating system with enough demand tosupport multiple applications in eachproduct category, he will be less likely tofind himself straitened later by having touse an application whose features disap-point him. Finally, the average userknows that, generally speaking, applica-tions improve through successive ver-sions. He thus wants an operating sys-tem for which successive generations ofhis favorite applications will be re-leased—promptly at that. The fact thata vastly larger number of applicationsare written for Windows than for otherPC operating systems attracts consum-ers to Windows, because it reassuresthem that their interests will be met aslong as they use Microsoft’s product.

Findings of Fact ¶ 37. Thus, despite thelimited success of its rivals, Microsoft ben-efits from the applications barrier to entry.

Of course, were middleware to succeed,it would erode the applications barrier toentry. Because applications written formultiple operating systems could run onany operating system on which the middle-ware product was present with little, ifany, porting, the operating system marketwould become competitive. Id. ¶ ¶ 29, 72.But as the District Court found, middle-ware will not expose a sufficient number ofAPIs to erode the applications barrier toentry in the foreseeable future. See id.¶ ¶ 28–29.


Microsoft next argues that the applica-tions barrier to entry is not an entry barri-er at all, but a reflection of Windows’popularity. It is certainly true that Win-dows may have gained its initial dominancein the operating system market competi-tively—through superior foresight or qual-ity. But this case is not about Microsoft’sinitial acquisition of monopoly power. It isabout Microsoft’s efforts to maintain thisposition through means other than compe-tition on the merits. Because the applica-tions barrier to entry protects a dominantoperating system irrespective of quality, itgives Microsoft power to stave off evensuperior new rivals. The barrier is thus acharacteristic of the operating system mar-ket, not of Microsoft’s popularity, or, asasserted by a Microsoft witness, the com-pany’s efficiency. See Direct Testimony ofRichard Schmalensee ¶ 115, reprinted in25 J.A. at 16153–14.

Finally, Microsoft argues that the Dis-trict Court should not have considered theapplications barrier to entry because itreflects not a cost borne disproportionate-ly by new entrants, but one borne by allparticipants in the operating system mar-ket. According to Microsoft, it had tomake major investments to convince soft-ware developers to write for its new oper-ating system, and it continues to ‘‘evan-gelize’’ the Windows platform today.Whether costs borne by all market partic-ipants should be considered entry barriersis the subject of much debate. Compare2A AREEDA & HOVENKAMP, ANTITRUST LAW

§ 420c, at 61 (arguing that these costs areentry barriers), and JOE S. BAIN, BARRIERS

TO NEW COMPETITION: THEIR CHARACTER AND

CONSEQUENCES IN MANUFACTURING INDUS-

TRIES 6–7 (1956) (considering these costsentry barriers), with L.A. Land Co. v.Brunswick Corp., 6 F.3d 1422, 1428 (9thCir.1993) (evaluating cost based on ‘‘[t]hedisadvantage of new entrants as comparedto incumbents’’), and GEORGE STIGLER, THE

ORGANIZATION OF INDUSTRY 67 (1968) (ex-cluding these costs). We need not resolvethis issue, however, for even under themore narrow definition it is clear thatthere are barriers. When Microsoft en-tered the operating system market withMS–DOS and the first version of Win-dows, it did not confront a dominant rivaloperating system with as massive an in-stalled base and as vast an existing arrayof applications as the Windows operatingsystems have since enjoyed. Findings ofFact ¶ ¶ 6, 7, 43. Moreover, when Micro-soft introduced Windows 95 and 98, it wasable to bypass the applications barrier toentry that protected the incumbent Win-dows by including APIs from the earlierversion in the new operating systems.See id. ¶ 44. This made porting existingWindows applications to the new versionof Windows much less costly than portingthem to the operating systems of otherentrants who could not freely includeAPIs from the incumbent Windows withtheir own.

2. Direct Proof

[11] Having sustained the DistrictCourt’s conclusion that circumstantial evi-dence proves that Microsoft possesses mo-nopoly power, we turn to Microsoft’s alter-native argument that it does not behavelike a monopolist. Claiming that softwarecompetition is uniquely ‘‘dynamic,’’ Appel-lant’s Opening Br. at 84 (quoting Findingsof Fact ¶ 59), the company suggests a newrule: that monopoly power in the softwareindustry should be proven directly, that is,by examining a company’s actual behaviorto determine if it reveals the existence ofmonopoly power. According to Microsoft,not only does no such proof of its powerexist, but record evidence demonstratesthe absence of monopoly power. The com-pany claims that it invests heavily in re-search and development, id. at 88–89 (cit-


ing Direct Testimony of Paul Maritz ¶ 155,reprinted in 6 J.A. at 3698 (testifying thatMicrosoft invests approximately 17% of itsrevenue in R&D)), and charges a low pricefor Windows (a small percentage of theprice of an Intel-compatible PC systemand less than the price of its rivals, id. at90 (citing Findings of Fact ¶ ¶ 19, 21, 46)).

Microsoft’s argument fails because, evenassuming that the software market isuniquely dynamic in the long term, theDistrict Court correctly applied the struc-tural approach to determine if the compa-ny faces competition in the short term.Structural market power analyses aremeant to determine whether potential sub-stitutes constrain a firm’s ability to raiseprices above the competitive level; onlythreats that are likely to materialize in therelatively near future perform this functionto any significant degree. Rothery, 792F.2d at 218 (quoting LAWRENCE SULLIVAN,

ANTITRUST § 12, at 41 (1977)) (only substi-tutes that can enter the market ‘‘prompt-ly’’ should be considered). The DistrictCourt expressly considered and rejectedMicrosoft’s claims that innovations such ashandheld devices and portal websiteswould soon expand the relevant marketbeyond Intel-compatible PC operating sys-tems. Because the company does notchallenge these findings, we have no rea-son to believe that prompt substitutes areavailable. The structural approach, as ap-plied by the District Court, is thus capableof fulfilling its purpose even in a changingmarket. Microsoft cites no case, nor arewe aware of one, requiring direct evidenceto show monopoly power in any market.We decline to adopt such a rule now.

Even if we were to require direct proof,moreover, Microsoft’s behavior may wellbe sufficient to show the existence of mo-nopoly power. Certainly, none of the con-duct Microsoft points to—its investment inR&D and the relatively low price of Win-

dows—is inconsistent with the possessionof such power. Conclusions of Law, at 37.The R&D expenditures Microsoft points toare not simply for Windows, but for itsentire company, which most likely does notpossess a monopoly for all of its products.Moreover, because innovation can increasean already dominant market share andfurther delay the emergence of competi-tion, even monopolists have reason to in-vest in R&D. Findings of Fact ¶ 61. Mi-crosoft’s pricing behavior is similarlyequivocal. The company claims only thatit never charged the short-term profit-maximizing price for Windows. Facedwith conflicting expert testimony, the Dis-trict Court found that it could not accu-rately determine what this price would be.Id. ¶ 65. In any event, the court found, aprice lower than the short-term profit-maximizing price is not inconsistent withpossession or improper use of monopolypower. Id. ¶ ¶ 65–66. Cf. Berkey Photo,Inc. v. Eastman Kodak Co., 603 F.2d 263,274 (2d Cir.1979) (‘‘[I]f monopoly powerhas been acquired or maintained throughimproper means, the fact that the powerhas not been used to extract [a monopolyprice] provides no succor to the monopo-list.’’). Microsoft never claims that it didnot charge the long-term monopoly price.Micosoft does argue that the price of Win-dows is a fraction of the price of an Intel-compatible PC system and lower than thatof rival operating systems, but these factsare not inconsistent with the DistrictCourt’s finding that Microsoft has monopo-ly power. See Findings of Fact ¶ 36 (‘‘In-tel-compatible PC operating systems otherthan Windows [would not] attract[ ] signifi-cant demand TTT even if Micosoft held itsprices substantially above the competitivelevel.’’).

More telling, the District Court foundthat some aspects of Microsoft’s behaviorare difficult to explain unless Windows is amonopoly product. For instance, accord-


ing to the District Court, the company setthe price of Windows without consideringrivals’ prices, Findings of Fact ¶ 62, some-thing a firm without a monopoly wouldhave been unable to do. The DistrictCourt also found that Microsoft’s patternof exclusionary conduct could only be ra-tional ‘‘if the firm knew that it possessedmonopoly power.’’ Conclusions of Law, at37. It is to that conduct that we now turn.

B. Anticompetitive Conduct

[12] As discussed above, having a mo-nopoly does not by itself violate § 2. Afirm violates § 2 only when it acquires ormaintains, or attempts to acquire or main-tain, a monopoly by engaging in exclusion-ary conduct ‘‘as distinguished from growthor development as a consequence of a su-perior product, business acumen, or histor-ic accident.’’ Grinnell, 384 U.S. at 571, 86S.Ct. 1698; see also United States v. Alu-minum Co. of Am., 148 F.2d 416, 430 (2dCir.1945) (Hand, J.) (‘‘The successful com-petitor, having been urged to compete,must not be turned upon when he wins.’’).

In this case, after concluding that Micro-soft had monopoly power, the DistrictCourt held that Microsoft had violated § 2by engaging in a variety of exclusionaryacts (not including predatory pricing), tomaintain its monopoly by preventing theeffective distribution and use of productsthat might threaten that monopoly. Spe-cifically, the District Court held Microsoftliable for: (1) the way in which it integrat-ed IE into Windows; (2) its various deal-ings with Original Equipment Manufactur-ers (‘‘OEMs’’), Internet Access Providers(‘‘IAPs’’), Internet Content Providers(‘‘ICPs’’), Independent Software Vendors(‘‘ISVs’’), and Apple Computer; (3) its ef-forts to contain and to subvert Java tech-nologies; and (4) its course of conduct as awhole. Upon appeal, Microsoft argues

that it did not engage in any exclusionaryconduct.

Whether any particular act of a monopo-list is exclusionary, rather than merely aform of vigorous competition, can be diffi-cult to discern: the means of illicit exclu-sion, like the means of legitimate competi-tion, are myriad. The challenge for anantitrust court lies in stating a generalrule for distinguishing between exclusion-ary acts, which reduce social welfare, andcompetitive acts, which increase it.

[13–15] From a century of case law onmonopolization under § 2, however, sever-al principles do emerge. First, to be con-demned as exclusionary, a monopolist’s actmust have an ‘‘anticompetitive effect.’’That is, it must harm the competitive pro-cess and thereby harm consumers. Incontrast, harm to one or more competitorswill not suffice. ‘‘The [Sherman Act] di-rects itself not against conduct which iscompetitive, even severely so, but againstconduct which unfairly tends to destroycompetition itself.’’ Spectrum Sports, Inc.v. McQuillan, 506 U.S. 447, 458, 113 S.Ct.884, 122 L.Ed.2d 247 (1993); see alsoBrooke Group Ltd. v. Brown & William-son Tobacco Corp., 509 U.S. 209, 225, 113S.Ct. 2578, 125 L.Ed.2d 168 (1993) (‘‘Evenan act of pure malice by one businesscompetitor against another does not, with-out more, state a claim under the federalantitrust lawsTTTT’’).

[16, 17] Second, the plaintiff, on whomthe burden of proof of course rests, see,e.g., Monsanto Co. v. Spray–Rite Serv.Corp., 465 U.S. 752, 763, 104 S.Ct. 1464, 79L.Ed.2d 775 (1984); see also United Statesv. Arnold, Schwinn & Co., 388 U.S. 365,374 n. 5, 87 S.Ct. 1856, 18 L.Ed.2d 1249(1967), overruled on other grounds, Cont’lT.V., Inc. v. GTE Sylvania Inc., 433 U.S.36, 97 S.Ct. 2549, 53 L.Ed.2d 568 (1977),must demonstrate that the monopolist’sconduct indeed has the requisite anticom-


petitive effect. See generally BrookeGroup, 509 U.S. at 225–26, 113 S.Ct. 2578.In a case brought by a private plaintiff, theplaintiff must show that its injury is ‘‘of‘the type that the statute was intended toforestall,’ ’’ Brunswick Corp. v. PuebloBowl–O–Mat, Inc., 429 U.S. 477, 487–88,97 S.Ct. 690, 50 L.Ed.2d 701 (1977) (quot-ing Wyandotte Transp. v. United States,389 U.S. 191, 202, 88 S.Ct. 379, 19 L.Ed.2d407 (1967)); no less in a case brought bythe Government, it must demonstrate thatthe monopolist’s conduct harmed competi-tion, not just a competitor.

[18, 19] Third, if a plaintiff successfullyestablishes a prima facie case under § 2by demonstrating anticompetitive effect,then the monopolist may proffer a ‘‘pro-competitive justification’’ for its conduct.See Eastman Kodak, 504 U.S. at 483, 112S.Ct. 2072. If the monopolist asserts aprocompetitive justification—a nonpretex-tual claim that its conduct is indeed a formof competition on the merits because itinvolves, for example, greater efficiency orenhanced consumer appeal—then the bur-den shifts back to the plaintiff to rebutthat claim. Cf. Capital Imaging Assocs.,P.C. v. Mohawk Valley Med. Assocs., Inc.,996 F.2d 537, 543 (2d Cir.1993).

[20] Fourth, if the monopolist’s pro-competitive justification stands unrebut-ted, then the plaintiff must demonstratethat the anticompetitive harm of the con-duct outweighs the procompetitive benefit.In cases arising under § 1 of the ShermanAct, the courts routinely apply a similarbalancing approach under the rubric of the‘‘rule of reason.’’ The source of the rule ofreason is Standard Oil Co. v. UnitedStates, 221 U.S. 1, 31 S.Ct. 502, 55 L.Ed.619 (1911), in which the Supreme Courtused that term to describe the proper in-quiry under both sections of the Act. Seeid. at 61–62, 31 S.Ct. 502 (‘‘[W]hen thesecond section [of the Sherman Act] is

thus harmonized with TTT the first, it be-comes obvious that the criteria to be re-sorted to in any given case for the purposeof ascertaining whether violations of thesection have been committed, is the rule ofreason guided by the establishedlawTTTT’’). As the Fifth Circuit more re-cently explained, ‘‘[i]t is clear TTT that theanalysis under section 2 is similar to thatunder section 1 regardless whether therule of reason label is appliedTTTT’’ Mid-Texas Communications Sys., Inc. v. AT &T, 615 F.2d 1372, 1389 n. 13 (5th Cir.1980)(citing Byars v. Bluff City News Co., 609F.2d 843, 860 (6th Cir.1979)); see also Cal.Computer Prods., Inc. v. IBM Corp., 613F.2d 727, 737 (9th Cir.1979).

[21] Finally, in considering whetherthe monopolist’s conduct on balance harmscompetition and is therefore condemned asexclusionary for purposes of § 2, our focusis upon the effect of that conduct, not uponthe intent behind it. Evidence of the in-tent behind the conduct of a monopolist isrelevant only to the extent it helps usunderstand the likely effect of the monopo-list’s conduct. See, e.g., Chicago Bd. ofTrade v. United States, 246 U.S. 231, 238,38 S.Ct. 242, 62 L.Ed. 683 (1918) (‘‘knowl-edge of intent may help the court to inter-pret facts and to predict consequences’’);Aspen Skiing Co. v. Aspen Highlands Ski-ing Corp., 472 U.S. 585, 603, 105 S.Ct.2847, 86 L.Ed.2d 467 (1985).

With these principles in mind, we nowconsider Microsoft’s objections to the Dis-trict Court’s holding that Microsoft violat-ed § 2 of the Sherman Act in a variety ofways.

1. Licenses Issued to Original Equip-ment Manufacturers

The District Court condemned a numberof provisions in Microsoft’s agreements li-censing Windows to OEMs, because it


found that Microsoft’s imposition of thoseprovisions (like many of Microsoft’s otheractions at issue in this case) serves toreduce usage share of Netscape’s browserand, hence, protect Microsoft’s operatingsystem monopoly. The reason marketshare in the browser market affects mar-ket power in the operating system marketis complex, and warrants some explana-tion.

Browser usage share is important be-cause, as we explained in Section II.Aabove, a browser (or any middleware prod-uct, for that matter) must have a criticalmass of users in order to attract softwaredevelopers to write applications relyingupon the APIs it exposes, and away fromthe APIs exposed by Windows. Applica-tions written to a particular browser’sAPIs, however, would run on any comput-er with that browser, regardless of theunderlying operating system. ‘‘The over-whelming majority of consumers will onlyuse a PC operating system for which therealready exists a large and varied set of TTT

applications, and for which it seems rela-tively certain that new types of applica-tions and new versions of existing applica-tions will continue to be marketedTTTT’’Findings of Fact ¶ 30. If a consumercould have access to the applications hedesired—regardless of the operating sys-tem he uses—simply by installing a partic-ular browser on his computer, then hewould no longer feel compelled to selectWindows in order to have access to thoseapplications; he could select an operatingsystem other than Windows based solelyupon its quality and price. In otherwords, the market for operating systemswould be competitive.

Therefore, Microsoft’s efforts to gainmarket share in one market (browsers)served to meet the threat to Microsoft’smonopoly in another market (operatingsystems) by keeping rival browsers from

gaining the critical mass of users neces-sary to attract developer attention awayfrom Windows as the platform for softwaredevelopment. Plaintiffs also argue thatMicrosoft’s actions injured competition inthe browser market—an argument we willexamine below in relation to their specificclaims that Microsoft attempted to monop-olize the browser market and unlawfullytied its browser to its operating system soas to foreclose competition in the browsermarket. In evaluating the § 2 monopolymaintenance claim, however, our immedi-ate concern is with the anticompetitive ef-fect of Microsoft’s conduct in preservingits monopoly in the operating system mar-ket.

In evaluating the restrictions in Micro-soft’s agreements licensing Windows toOEMs, we first consider whether plaintiffshave made out a prima facie case by dem-onstrating that the restrictions have ananticompetitive effect. In the next subsec-tion, we conclude that plaintiffs have metthis burden as to all the restrictions. Wethen consider Microsoft’s proffered justifi-cations for the restrictions and, for themost part, hold those justifications insuffi-cient.

a. Anticompetitive effect of the licenserestrictions

[22] The restrictions Microsoft placesupon Original Equipment Manufacturersare of particular importance in determin-ing browser usage share because havingan OEM pre-install a browser on a com-puter is one of the two most cost-effectivemethods by far of distributing browsingsoftware. (The other is bundling thebrowser with internet access software dis-tributed by an IAP.) Findings of Fact¶ 145. The District Court found that therestrictions Microsoft imposed in licensingWindows to OEMs prevented many OEMsfrom distributing browsers other than IE.


Conclusions of Law, at 39–40. In particu-lar, the District Court condemned the li-cense provisions prohibiting the OEMsfrom: (1) removing any desktop icons,folders, or ‘‘Start’’ menu entries; (2) alter-ing the initial boot sequence; and (3) oth-erwise altering the appearance of the Win-dows desktop. Findings of Fact ¶ 213.

The District Court concluded that thefirst license restriction—the prohibitionupon the removal of desktop icons, folders,and Start menu entries—thwarts the dis-tribution of a rival browser by preventingOEMs from removing visible means ofuser access to IE. Id. ¶ 203. The OEMscannot practically install a second browserin addition to IE, the court found, in partbecause ‘‘[p]re-installing more than oneproduct in a given category TTT can signifi-cantly increase an OEM’s support costs,for the redundancy can lead to confusionamong novice users.’’ Id. ¶ 159; see alsoid. ¶ 217. That is, a certain number ofnovice computer users, seeing two browsericons, will wonder which to use when andwill call the OEM’s support line. Supportcalls are extremely expensive and, in thehighly competitive original equipment mar-ket, firms have a strong incentive to mini-mize costs. Id. ¶ 210.

Microsoft denies the ‘‘consumer confu-sion’’ story; it observes that some OEMsdo install multiple browsers and that exec-utives from two OEMs that do so deniedany knowledge of consumers being con-fused by multiple icons. See 11/5/98 pmTr. at 41–42 (trial testimony of AvadisTevanian of Apple), reprinted in 9 J.A. at5493–94; 11/18/99 am Tr. at 69 (trial testi-mony of John Soyring of IBM), reprintedin 10 J.A. at 6222.

Other testimony, however, supports theDistrict Court’s finding that fear of suchconfusion deters many OEMs from pre-installing multiple browsers. See, e.g.,01/13/99 pm Tr. at 614–15 (deposition of

Microsoft’s Gayle McClain played to thecourt) (explaining that redundancy of iconsmay be confusing to end users); 02/18/99pm Tr. at 46–47 (trial testimony of JohnRose of Compaq), reprinted in 21 J.A. at14237–38 (same); 11/17/98 am Tr. at 68(deposition of John Kies of Packard Bell–NEC played to the court), reprinted in 9J.A. at 6016 (same); 11/17/98 am Tr. at 67–72 (trial testimony of Glenn Weadock), re-printed in 9 J.A. at 6015–20 (same). Mosttelling, in presentations to OEMs, Micro-soft itself represented that having only oneicon in a particular category would be ‘‘lessconfusing for endusers.’’ See Govern-ment’s Trial Exhibit (‘‘GX’’) 319 at MS980109453. Accordingly, we reject Micro-soft’s argument that we should vacate theDistrict Court’s Finding of Fact 159 as itrelates to consumer confusion.

As noted above, the OEM channel is oneof the two primary channels for distribu-tion of browsers. By preventing OEMsfrom removing visible means of user ac-cess to IE, the license restriction preventsmany OEMs from pre-installing a rivalbrowser and, therefore, protects Micro-soft’s monopoly from the competition thatmiddleware might otherwise present.Therefore, we conclude that the licenserestriction at issue is anticompetitive. Wedefer for the moment the question whetherthat anticompetitive effect is outweighedby Microsoft’s proffered justifications.

The second license provision at issueprohibits OEMs from modifying the initialboot sequence—the process that occursthe first time a consumer turns on thecomputer. Prior to the imposition of thatrestriction, ‘‘among the programs thatmany OEMs inserted into the boot se-quence were Internet sign-up proceduresthat encouraged users to choose from a listof IAPs assembled by the OEM.’’ Find-ings of Fact ¶ 210. Microsoft’s prohibitionon any alteration of the boot sequence thus


prevents OEMs from using that process topromote the services of IAPs, many ofwhich—at least at the time Microsoft im-posed the restriction—used Navigatorrather than IE in their internet accesssoftware. See id. ¶ 212; GX 295, reprintedin 12 J.A. at 14533 (Upon learning of OEMpractices including boot sequence modifica-tion, Microsoft’s Chairman, Bill Gates,wrote: ‘‘Apparently a lot of OEMs arebundling non-Microsoft browsers and com-ing up with offerings together with [IAPs]that get displayed on their machines in aFAR more prominent way than MSN orour Internet browser.’’). Microsoft doesnot deny that the prohibition on modifyingthe boot sequence has the effect of de-creasing competition against IE by pre-venting OEMs from promoting rivals’browsers. Because this prohibition has asubstantial effect in protecting Microsoft’smarket power, and does so through ameans other than competition on the mer-its, it is anticompetitive. Again the ques-tion whether the provision is nonethelessjustified awaits later treatment.

Finally, Microsoft imposes several addi-tional provisions that, like the prohibitionon removal of icons, prevent OEMs frommaking various alterations to the desktop:Microsoft prohibits OEMs from causingany user interface other than the Windowsdesktop to launch automatically, from add-ing icons or folders different in size orshape from those supplied by Microsoft,and from using the ‘‘Active Desktop’’ fea-ture to promote third-party brands.These restrictions impose significant costsupon the OEMs; prior to Microsoft’s pro-hibiting the practice, many OEMs wouldchange the appearance of the desktop inways they found beneficial. See, e.g.,Findings of Fact ¶ 214; GX 309, reprintedin 22 J.A. at 14551 (March 1997 letterfrom Hewlett–Packard to Microsoft: ‘‘Weare responsible for the cost of technicalsupport of our customers, including the

33% of calls we get related to the lack ofquality or confusion generated by yourproductTTTT We must have more abilityto decide how our system is presented toour end users. If we had a choice ofanother supplier, based on your actions inthis area, I assure you [that you] would notbe our supplier of choice.’’).

The dissatisfaction of the OEM custom-ers does not, of course, mean the restric-tions are anticompetitive. The anticom-petitive effect of the license restrictions is,as Microsoft itself recognizes, that OEMsare not able to promote rival browsers,which keeps developers focused upon theAPIs in Windows. Findings of Fact ¶ 212(quoting Microsoft’s Gates as writing,‘‘[w]inning Internet browser share is avery very important goal for us,’’ and em-phasizing the need to prevent OEMs frompromoting both rival browsers and IAPsthat might use rivals’ browsers); see also01/13/99 Tr. at 305–06 (excerpts from de-position of James Von Holle of Gateway)(prior to restriction Gateway had pre-in-stalled non-IE internet registration iconthat was larger than other desktop icons).This kind of promotion is not a zero-sumgame; but for the restrictions in theirlicenses to use Windows, OEMs could pro-mote multiple IAPs and browsers. Bypreventing the OEMs from doing so, thistype of license restriction, like the first tworestrictions, is anticompetitive: Microsoftreduced rival browsers’ usage share not byimproving its own product but, rather, bypreventing OEMs from taking actions thatcould increase rivals’ share of usage.

b. Microsoft’s justifications for the li-cense restrictions

Microsoft argues that the license re-strictions are legally justified because, inimposing them, Microsoft is simply ‘‘exer-cising its rights as the holder of validcopyrights.’’ Appellant’s Opening Br. at


102. Microsoft also argues that the licens-es ‘‘do not unduly restrict the opportuni-ties of Netscape to distribute Navigator inany event.’’ Id.

[23] Microsoft’s primary copyright ar-gument borders upon the frivolous. Thecompany claims an absolute and unfetteredright to use its intellectual property as itwishes: ‘‘[I]f intellectual property rightshave been lawfully acquired,’’ it says, then‘‘their subsequent exercise cannot give riseto antitrust liability.’’ Appellant’s OpeningBr. at 105. That is no more correct thanthe proposition that use of one’s personalproperty, such as a baseball bat, cannotgive rise to tort liability. As the FederalCircuit succinctly stated: ‘‘Intellectualproperty rights do not confer a privilege toviolate the antitrust laws.’’ In re Indep.Serv. Orgs. Antitrust Litig., 203 F.3d 1322,1325 (Fed.Cir.2000).

Although Microsoft never overtly re-treats from its bold and incorrect positionon the law, it also makes two arguments tothe effect that it is not exercising its copy-right in an unreasonable manner, despitethe anticompetitive consequences of thelicense restrictions discussed above. Inthe first variation upon its unqualifiedcopyright defense, Microsoft cites twocases indicating that a copyright holdermay limit a licensee’s ability to engage insignificant and deleterious alterations of acopyrighted work. See Gilliam v. ABC,538 F.2d 14, 21 (2d Cir.1976); WGN Cont’lBroad. Co. v. United Video, Inc., 693 F.2d622, 625 (7th Cir.1982). The relevance ofthose two cases for the present one islimited, however, both because those casesinvolved substantial alterations of a copy-righted work, see Gilliam, 538 F.2d at 18,and because in neither case was there anyclaim that the copyright holder was, inasserting its rights, violating the antitrustlaws, see WGN Cont’l Broad., 693 F.2d at626; see also Cmty. for Creative Non–

Violence v. Reid, 846 F.2d 1485, 1498(D.C.Cir.1988) (noting, again in a contextfree of any antitrust concern, that ‘‘anauthor [ ] may have rights against’’ a licen-see that ‘‘excessively mutilated or altered’’the copyrighted work).

[24] The only license restriction Micro-soft seriously defends as necessary to pre-vent a ‘‘substantial alteration’’ of its copy-righted work is the prohibition on OEMsautomatically launching a substitute userinterface upon completion of the boot pro-cess. See Findings of Fact ¶ 211 (‘‘[A] fewlarge OEMs developed programs that ranautomatically at the conclusion of a newPC system’s first boot sequence. Theseprograms replaced the Windows desktopeither with a user interface designed bythe OEM or with Navigator’s user inter-face.’’). We agree that a shell that auto-matically prevents the Windows desktopfrom ever being seen by the user is adrastic alteration of Microsoft’s copyright-ed work, and outweighs the marginal anti-competitive effect of prohibiting the OEMsfrom substituting a different interface au-tomatically upon completion of the initialboot process. We therefore hold that thisparticular restriction is not an exclusionarypractice that violates § 2 of the ShermanAct.

[25] In a second variation upon itscopyright defense, Microsoft argues thatthe license restrictions merely preventOEMs from taking actions that would re-duce substantially the value of Microsoft’scopyrighted work: that is, Microsoftclaims each license restriction in questionis necessary to prevent OEMs from soaltering Windows as to undermine ‘‘theprincipal value of Windows as a stable andconsistent platform that supports a broadrange of applications and that is familiar tousers.’’ Appellant’s Opening Br. at 102.Microsoft, however, never substantiatesthis claim, and, because an OEM’s altering


the appearance of the desktop or promot-ing programs in the boot sequence doesnot affect the code already in the product,the practice does not self-evidently affecteither the ‘‘stability’’ or the ‘‘consistency’’of the platform. See Conclusions of Law,at 41; Findings of Fact ¶ 227. Microsoftcites only one item of evidence in supportof its claim that the OEMs’ alterationswere decreasing the value of Windows.Defendant’s Trial Exhibit (‘‘DX’’) 2395 atMSV0009378A, reprinted in 19 J.A. at12575. That document, prepared by Mi-crosoft itself, states: ‘‘there are qualityissues created by OEMs who are too lib-eral with the pre-install process,’’ referringto the OEMs’ installation of Windows andadditional software on their PCs, which thedocument says may result in ‘‘user con-cerns and confusion.’’ To the extent theOEMs’ modifications cause consumer con-fusion, of course, the OEMs bear the addi-tional support costs. See Findings of Fact¶ 159. Therefore, we conclude Microsofthas not shown that the OEMs’ liberalityreduces the value of Windows except inthe sense that their promotion of rivalbrowsers undermines Microsoft’s monopo-ly—and that is not a permissible justifica-tion for the license restrictions.

[26] Apart from copyright, Microsoftraises one other defense of the OEM li-cense agreements: It argues that, despitethe restrictions in the OEM license, Net-scape is not completely blocked from dis-tributing its product. That claim is insuffi-cient to shield Microsoft from liability forthose restrictions because, although Micro-soft did not bar its rivals from all means ofdistribution, it did bar them from the cost-efficient ones.

In sum, we hold that with the exceptionof the one restriction prohibiting automati-cally launched alternative interfaces, allthe OEM license restrictions at issue rep-resent uses of Microsoft’s market power to

protect its monopoly, unredeemed by anylegitimate justification. The restrictionstherefore violate § 2 of the Sherman Act.

2. Integration of IE and Windows

Although Microsoft’s license restrictionshave a significant effect in closing rivalbrowsers out of one of the two primarychannels of distribution, the District Courtfound that ‘‘Microsoft’s executives believedTTT its contractual restrictions placed onOEMs would not be sufficient in them-selves to reverse the direction of Naviga-tor’s usage share. Consequently, in late1995 or early 1996, Microsoft set out tobind [IE] more tightly to Windows 95 as atechnical matter.’’ Findings of Fact ¶ 160.

Technologically binding IE to Windows,the District Court found, both preventedOEMs from pre-installing other browsersand deterred consumers from using them.In particular, having the IE software codeas an irremovable part of Windows meantthat pre-installing a second browser would‘‘increase an OEM’s product testing costs,’’because an OEM must test and train itssupport staff to answer calls related toevery software product preinstalled on themachine; moreover, pre-installing a brow-ser in addition to IE would to many OEMsbe ‘‘a questionable use of the scarce andvaluable space on a PC’s hard drive.’’ Id.¶ 159.

Although the District Court, in its Con-clusions of Law, broadly condemned Mi-crosoft’s decision to bind ‘‘Internet Ex-plorer to Windows with TTT technologicalshackles,’’ Conclusions of Law, at 39, itsfindings of fact in support of that conclu-sion center upon three specific actionsMicrosoft took to weld IE to Windows:excluding IE from the ‘‘Add/Remove Pro-grams’’ utility; designing Windows so asin certain circumstances to override theuser’s choice of a default browser otherthan IE; and commingling code related


to browsing and other code in the samefiles, so that any attempt to delete thefiles containing IE would, at the sametime, cripple the operating system. Aswith the license restrictions, we considerfirst whether the suspect actions had ananticompetitive effect, and then whetherMicrosoft has provided a procompetitivejustification for them.

a. Anticompetitive effect of integration

[27] As a general rule, courts are prop-erly very skeptical about claims that com-petition has been harmed by a dominantfirm’s product design changes. See, e.g.,Foremost Pro Color, Inc. v. Eastman Ko-dak Co., 703 F.2d 534, 544–45 (9th Cir.1983). In a competitive market, firms rou-tinely innovate in the hope of appealing toconsumers, sometimes in the process mak-ing their products incompatible with thoseof rivals; the imposition of liability when amonopolist does the same thing will inevi-tably deter a certain amount of innovation.This is all the more true in a market, suchas this one, in which the product itself israpidly changing. See Findings of Fact¶ 59. Judicial deference to product inno-vation, however, does not mean that amonopolist’s product design decisions areper se lawful. See Foremost Pro Color,703 F.2d at 545; see also Cal. ComputerProds., 613 F.2d at 739, 744; In re IBMPeripheral EDP Devices Antitrust Litig.,481 F.Supp. 965, 1007–08 (N.D.Cal.1979).

[28] The District Court first con-demned as anticompetitive Microsoft’s de-cision to exclude IE from the ‘‘Add/Re-move Programs’’ utility in Windows 98.Findings of Fact ¶ 170. Microsoft hadincluded IE in the Add/Remove Programsutility in Windows 95, see id. ¶ ¶ 175–76,but when it modified Windows 95 to pro-duce Windows 98, it took IE out of theAdd/Remove Programs utility. Thischange reduces the usage share of rival

browsers not by making Microsoft’s ownbrowser more attractive to consumers but,rather, by discouraging OEMs from dis-tributing rival products. See id. ¶ 159.Because Microsoft’s conduct, throughsomething other than competition on themerits, has the effect of significantly re-ducing usage of rivals’ products and henceprotecting its own operating system mo-nopoly, it is anticompetitive; we defer forthe moment the question whether it isnonetheless justified.

Second, the District Court found thatMicrosoft designed Windows 98 ‘‘so thatusing Navigator on Windows 98 wouldhave unpleasant consequences for users’’by, in some circumstances, overriding theuser’s choice of a browser other than IE ashis or her default browser. Id. ¶ ¶ 171–72.Plaintiffs argue that this override harmsthe competitive process by deterring con-sumers from using a browser other thanIE even though they might prefer to do so,thereby reducing rival browsers’ usageshare and, hence, the ability of rival brow-sers to draw developer attention awayfrom the APIs exposed by Windows. Mi-crosoft does not deny, of course, that over-riding the user’s preference prevents somepeople from using other browsers. Be-cause the override reduces rivals’ usageshare and protects Microsoft’s monopoly, ittoo is anticompetitive.

[29] Finally, the District Court con-demned Microsoft’s decision to bind IE toWindows 98 ‘‘by placing code specific toWeb browsing in the same files as codethat provided operating system functions.’’Id. ¶ 161; see also id. ¶ ¶ 174, 192. Put-ting code supplying browsing functionalityinto a file with code supplying operatingsystem functionality ‘‘ensure[s] that thedeletion of any file containing browsing-specific routines would also delete vitaloperating system routines and thus crippleWindowsTTTT’’ Id. ¶ 164. As noted above,


preventing an OEM from removing IEdeters it from installing a second browserbecause doing so increases the OEM’sproduct testing and support costs; by con-trast, had OEMs been able to remove IE,they might have chosen to pre-install Navi-gator alone. See id. ¶ 159.

Microsoft denies, as a factual matter,that it commingled browsing and non-browsing code, and it maintains the Dis-trict Court’s findings to the contrary areclearly erroneous. According to Microsoft,its expert ‘‘testified without contradictionthat ‘[t]he very same code in Windows 98that provides Web browsing functionality’also performs essential operating systemfunctions—not code in the same files, butthe very same software code.’’ Appellant’sOpening Br. at 79 (citing 5 J.A. 3291–92).

Microsoft’s expert did not testify to thateffect ‘‘without contradiction,’’ however. AGovernment expert, Glenn Weadock, testi-fied that Microsoft ‘‘design[ed] [IE] so thatsome of the code that it uses co-resides inthe same library files as other code neededfor Windows.’’ Direct Testimony ¶ 30.Another Government expert likewise testi-fied that one library file, SHDOCVW.DLL,‘‘is really a bundle of separate functions.It contains some functions that have to dospecifically with Web browsing, and it con-tains some general user interface functionsas well.’’ 12/14/98 am Tr. at 60–61 (trialtestimony of Edward Felten), reprinted in11 J.A. at 6953–54. One of Microsoft’sown documents suggests as much. SeePlaintiffs’ Proposed Findings of Fact¶ 131.2.vii (citing GX 1686 (under seal) (Mi-crosoft document indicating some functionsin SHDOCVW.DLL can be described as‘‘IE only,’’ others can be described as‘‘shell only’’ and still others can be de-scribed as providing both ‘‘IE’’ and ‘‘shell’’functions)).

In view of the contradictory testimony inthe record, some of which supports the

District Court’s finding that Microsoftcommingled browsing and non-browsingcode, we cannot conclude that the findingwas clearly erroneous. See Anderson v.City of Bessemer City, 470 U.S. 564, 573–74, 105 S.Ct. 1504, 84 L.Ed.2d 518 (1985)(‘‘If the district court’s account of the evi-dence is plausible in light of the recordviewed in its entirety, the court of appealsmay not reverse it even though convincedthat had it been sitting as the trier of fact,it would have weighed the evidence differ-ently.’’). Accordingly, we reject Micro-soft’s argument that we should vacateFinding of Fact 159 as it relates to thecommingling of code, and we conclude thatsuch commingling has an anticompetitiveeffect; as noted above, the comminglingdeters OEMs from pre-installing rivalbrowsers, thereby reducing the rivals’ us-age share and, hence, developers’ interestin rivals’ APIs as an alternative to the APIset exposed by Microsoft’s operating sys-tem.

b. Microsoft’s justifications for inte-gration

[30] Microsoft proffers no justificationfor two of the three challenged actionsthat it took in integrating IE into Win-dows—excluding IE from the Add/Re-move Programs utility and comminglingbrowser and operating system code. Al-though Microsoft does make some generalclaims regarding the benefits of integrat-ing the browser and the operating system,see, e.g., Direct Testimony of James All-chin ¶ 94, reprinted in 5 J.A. at 3321(‘‘Our vision of deeper levels of technicalintegration is highly efficient and providessubstantial benefits to customers and de-velopers.’’), it neither specifies nor sub-stantiates those claims. Nor does it ar-gue that either excluding IE from theAdd/Remove Programs utility or com-mingling code achieves any integrative


benefit. Plaintiffs plainly made out a pri-ma facie case of harm to competition inthe operating system market by demon-strating that Microsoft’s actions increasedits browser usage share and thus protect-ed its operating system monopoly from amiddleware threat and, for its part, Mi-crosoft failed to meet its burden of show-ing that its conduct serves a purpose oth-er than protecting its operating systemmonopoly. Accordingly, we hold that Mi-crosoft’s exclusion of IE from the Add/Re-move Programs utility and its comming-ling of browser and operating system codeconstitute exclusionary conduct, in viola-tion of § 2.

As for the other challenged act thatMicrosoft took in integrating IE into Win-dows—causing Windows to override theuser’s choice of a default browser in cer-tain circumstances—Microsoft argues thatit has ‘‘valid technical reasons.’’ Specifi-cally, Microsoft claims that it was neces-sary to design Windows to override theuser’s preferences when he or she invokesone of ‘‘a few’’ out ‘‘of the nearly 30 meansof accessing the Internet.’’ Appellant’sOpening Br. at 82. According to Micro-soft:

The Windows 98 Help system and Win-dows Update feature depend on ActiveXcontrols not supported by Navigator,and the now-discontinued Channel Barutilized Microsoft’s Channel DefinitionFormat, which Navigator also did notsupport. Lastly, Windows 98 does notinvoke Navigator if a user accesses theInternet through ‘‘My Computer’’ or‘‘Windows Explorer’’ because doing sowould defeat one of the purposes ofthose features—enabling users to moveseamlessly from local storage devices tothe Web in the same browsing window.

Id. (internal citations omitted). The plain-tiff bears the burden not only of rebuttinga proffered justification but also of demon-

strating that the anticompetitive effect ofthe challenged action outweighs it. In theDistrict Court, plaintiffs appear to havedone neither, let alone both; in any event,upon appeal, plaintiffs offer no rebuttalwhatsoever. Accordingly, Microsoft maynot be held liable for this aspect of itsproduct design.

3. Agreements with Internet AccessProviders

The District Court also condemned asexclusionary Microsoft’s agreements withvarious IAPs. The IAPs include both In-ternet Service Providers, which offer con-sumers internet access, and Online Ser-vices (‘‘OLSs’’) such as America Online(‘‘AOL’’), which offer proprietary contentin addition to internet access and otherservices. Findings of Fact ¶ 15. The Dis-trict Court deemed Microsoft’s agreementswith the IAPs unlawful because:

Microsoft licensed [IE] and the [IE] Ac-cess Kit [ (of which, more below) ] tohundreds of IAPs for no charge. [Find-ings of Fact] ¶ ¶ 250–51. Then, Micro-soft extended valuable promotionaltreatment to the ten most importantIAPs in exchange for their commitmentto promote and distribute [IE] and toexile Navigator from the desktop. Id.¶ ¶ 255–58, 261, 272, 288–90, 305–06. Fi-nally, in exchange for efforts to upgradeexisting subscribers to client softwarethat came bundled with [IE] instead ofNavigator, Microsoft granted rebates—and in some cases made outright pay-ments—to those same IAPs. Id. ¶ ¶ 259–60, 295.

Conclusions of Law, at 41.

[31] The District Court condemned Mi-crosoft’s actions in (1) offering IE free ofcharge to IAPs and (2) offering IAPs abounty for each customer the IAP signs upfor service using the IE browser. In ef-fect, the court concluded that Microsoft is


acting to preserve its monopoly by offeringIE to IAPs at an attractive price. Similar-ly, the District Court held Microsoft liablefor (3) developing the IE Access Kit(‘‘IEAK’’), a software package that allowsan IAP to ‘‘create a distinctive identity forits service in as little as a few hours bycustomizing the [IE] title bar, icon, startand search pages,’’ Findings of Fact ¶ 249,and (4) offering the IEAK to IAPs free ofcharge, on the ground that those acts, too,helped Microsoft preserve its monopoly.Conclusions of Law, at 41–42. Finally, theDistrict Court found that (5) Microsoftagreed to provide easy access to IAPs’services from the Windows desktop in re-turn for the IAPs’ agreement to promoteIE exclusively and to keep shipments ofinternet access software using Navigatorunder a specific percentage, typically 25%.See Conclusions of Law, at 42 (citingFindings of Fact ¶ ¶ 258, 262, 289). Weaddress the first four items—Microsoft’sinducements—and then its exclusiveagreements with IAPs.

Although offering a customer an attrac-tive deal is the hallmark of competition,the Supreme Court has indicated that invery rare circumstances a price may beunlawfully low, or ‘‘predatory.’’ See gener-ally Brooke Group, 509 U.S. at 220–27, 113S.Ct. 2578. Plaintiffs argued before theDistrict Court that Microsoft’s pricing wasindeed predatory; but instead of makingthe usual predatory pricing argument—that the predator would drive out its rivalsby pricing below cost on a particular prod-uct and then, sometime in the future, raiseits prices on that product above the com-petitive level in order to recoup its earlierlosses—plaintiffs argued that by pricingbelow cost on IE (indeed, even payingpeople to take it), Microsoft was able si-multaneously to preserve its stream of mo-nopoly profits on Windows, thereby morethan recouping its investment in below-cost pricing on IE. The District Court did

not assign liability for predatory pricing,however, and plaintiffs do not press thistheory on appeal.

[32] The rare case of price predationaside, the antitrust laws do not condemneven a monopolist for offering its productat an attractive price, and we thereforehave no warrant to condemn Microsoft foroffering either IE or the IEAK free ofcharge or even at a negative price. Like-wise, as we said above, a monopolist doesnot violate the Sherman Act simply bydeveloping an attractive product. SeeGrinnell, 384 U.S. at 571, 86 S.Ct. 1698(‘‘[G]rowth or development as a conse-quence of a superior product [or] businessacumen’’ is no violation.). Therefore, Mi-crosoft’s development of the IEAK doesnot violate the Sherman Act.

[33] We turn now to Microsoft’s dealswith IAPs concerning desktop placement.Microsoft concluded these exclusive agree-ments with all ‘‘the leading IAPs,’’ Find-ings of Fact ¶ 244, including the majorOLSs. Id. ¶ 245; see also id. ¶ ¶ 305, 306.The most significant of the OLS deals iswith AOL, which, when the deal wasreached, ‘‘accounted for a substantial por-tion of all existing Internet access sub-scriptions and TTT attracted a very largepercentage of new IAP subscribers.’’ Id.¶ 272. Under that agreement Microsoftputs the AOL icon in the OLS folder onthe Windows desktop and AOL does notpromote any non-Microsoft browser, norprovide software using any non-Microsoftbrowser except at the customer’s request,and even then AOL will not supply morethan 15% of its subscribers with a browserother than IE. Id. ¶ 289.

The Supreme Court most recently con-sidered an antitrust challenge to an exclu-sive contract in Tampa Electric Co. v.Nashville Coal Co., 365 U.S. 320, 81 S.Ct.623, 5 L.Ed.2d 580 (1961). That case,


which involved a challenge to a require-ments contract, was brought under § 3 ofthe Clayton Act and §§ 1 and 2 of theSherman Act. The Court held that anexclusive contract does not violate theClayton Act unless its probable effect is to‘‘foreclose competition in a substantialshare of the line of commerce affected.’’Id. at 327, 81 S.Ct. 623. The share of themarket foreclosed is important because,for the contract to have an adverse effectupon competition, ‘‘the opportunities forother traders to enter into or remain inthat market must be significantly limited.’’Id. at 328, 81 S.Ct. 623. Although ‘‘[n]ei-ther the Court of Appeals nor the DistrictCourt [had] considered in detail the ques-tion of the relevant market,’’ id. at 330, 81S.Ct. 623, the Court in Tampa Electricexamined the record and, after definingthe relevant market, determined that thecontract affected less than one percent ofthat market. Id. at 333, 81 S.Ct. 623.After concluding, under the Clayton Act,that this share was ‘‘conservatively speak-ing, quite insubstantial,’’ id., the Courtwent on summarily to reject the ShermanAct claims. Id. at 335, 81 S.Ct. 623 (‘‘[I]f[the contract] does not fall within thebroader prescription of § 3 of the ClaytonAct it follows that it is not forbidden bythose of the [Sherman Act].’’).

Following Tampa Electric, courts con-sidering antitrust challenges to exclusivecontracts have taken care to identify theshare of the market foreclosed. Somecourts have indicated that § 3 of the Clay-ton Act and § 1 of the Sherman Act re-quire an equal degree of foreclosure beforeprohibiting exclusive contracts. See, e.g.,Roland Mach. Co. v. Dresser Indus., Inc.,749 F.2d 380, 393 (7th Cir.1984) (Posner,J.). Other courts, however, have held thata higher market share must be foreclosedin order to establish a violation of theSherman Act as compared to the ClaytonAct. See, e.g., Barr Labs. v. Abbott Labs.,

978 F.2d 98, 110 (3d Cir.1992); 11 HER-

BERT HOVENKAMP, ANTITRUST LAW ¶ 1800c4(1998) (‘‘[T]he cases are divided, with alikely majority stating that the ClaytonAct requires a smaller showing of anticom-petitive effects.’’).

[34] Though what is ‘‘significant’’ mayvary depending upon the antitrust provi-sion under which an exclusive deal is chal-lenged, it is clear that in all cases theplaintiff must both define the relevantmarket and prove the degree of foreclo-sure. This is a prudential requirement;exclusivity provisions in contracts mayserve many useful purposes. See, e.g.,Omega Envtl., Inc. v. Gilbarco, Inc., 127F.3d 1157, 1162 (9th Cir.1997) (‘‘There are,however, well-recognized economic bene-fits to exclusive dealing arrangements, in-cluding the enhancement of interbrandcompetition.’’); Barry Wright Corp. v. ITTGrinnell Corp., 724 F.2d 227, 236 (1st Cir.1983) (Breyer, J.) (‘‘[V]irtually every con-tract to buy ‘forecloses’ or ‘excludes’ alter-native sellers from some portion of themarket, namely the portion consisting ofwhat was bought.’’). Permitting an anti-trust action to proceed any time a firmenters into an exclusive deal would bothdiscourage a presumptively legitimatebusiness practice and encourage costly an-titrust actions. Because an exclusive dealaffecting a small fraction of a marketclearly cannot have the requisite harmfuleffect upon competition, the requirementof a significant degree of foreclosureserves a useful screening function. Cf.Frank H. Easterbrook, The Limits of An-titrust, 63 TEX. L. REV. 1, 21–23 (1984)(discussing use of presumptions in anti-trust law to screen out cases in which lossto consumers and economy is likely out-weighed by cost of inquiry and risk ofdeterring procompetitive behavior).


In this case, plaintiffs challenged Micro-soft’s exclusive dealing arrangements withthe IAPs under both §§ 1 and 2 of theSherman Act. The District Court, in ana-lyzing the § 1 claim, stated, ‘‘unless theevidence demonstrates that Microsoft’sagreements excluded Netscape altogetherfrom access to roughly forty percent of thebrowser market, the Court should declineto find such agreements in violation of§ 1.’’ Conclusions of Law, at 52. Thecourt recognized that Microsoft had sub-stantially excluded Netscape from ‘‘themost efficient channels for Navigator toachieve browser usage share,’’ id. at 53;see also Findings of Fact ¶ 145 (‘‘[N]oother distribution channel for browsingsoftware even approaches the efficiency ofOEM pre-installation and IAP bundling.’’),and had relegated it to more costly andless effective methods (such as mass mail-ing its browser on a disk or offering it fordownload over the internet); but becauseMicrosoft has not ‘‘completely excludedNetscape’’ from reaching any potentialuser by some means of distribution, how-ever ineffective, the court concluded theagreements do not violate § 1. Conclu-sions of Law, at 53. Plaintiffs did notcross-appeal this holding.

Turning to § 2, the court stated: ‘‘thefact that Microsoft’s arrangements withvarious [IAPs and other] firms did notforeclose enough of the relevant market toconstitute a § 1 violation in no way de-tracts from the Court’s assignment of lia-bility for the same arrangements under§ 2TTTT [A]ll of Microsoft’s agreements,including the non-exclusive ones, severelyrestricted Netscape’s access to those dis-tribution channels leading most efficientlyto the acquisition of browser usage share.’’Conclusions of Law, at 53.

On appeal Microsoft argues that ‘‘courtshave applied the same standard to allegedexclusive dealing agreements under both

Section 1 and Section 2,’’ Appellant’sOpening Br. at 109, and it argues that theDistrict Court’s holding of no liability un-der § 1 necessarily precludes holding itliable under § 2. The District Court ap-pears to have based its holding with re-spect to § 1 upon a ‘‘total exclusion test’’rather than the 40% standard drawn fromthe caselaw. Even assuming the holdingis correct, however, we nonetheless rejectMicrosoft’s contention.

[35] The basic prudential concerns rel-evant to §§ 1 and 2 are admittedly thesame: exclusive contracts are common-place—particularly in the field of distribu-tion—in our competitive, market economy,and imposing upon a firm with marketpower the risk of an antitrust suit everytime it enters into such a contract, nomatter how small the effect, would createan unacceptable and unjustified burdenupon any such firm. At the same time,however, we agree with plaintiffs that amonopolist’s use of exclusive contracts, incertain circumstances, may give rise to a§ 2 violation even though the contractsforeclose less than the roughly 40% or 50%share usually required in order to establisha § 1 violation. See generally Dennis W.Carlton, A General Analysis of Exclusion-ary Conduct and Refusal to Deal—WhyAspen and Kodak Are Misguided, 68 ANTI-

TRUST L.J. 659 (2001) (explaining variousscenarios under which exclusive dealing,particularly by a dominant firm, may raiselegitimate concerns about harm to compe-tition).

In this case, plaintiffs allege that, byclosing to rivals a substantial percentageof the available opportunities for browserdistribution, Microsoft managed to pre-serve its monopoly in the market for oper-ating systems. The IAPs constitute one ofthe two major channels by which browserscan be distributed. Findings of Fact¶ 242. Microsoft has exclusive deals with


‘‘fourteen of the top fifteen access provid-ers in North America[, which] account fora large majority of all Internet access sub-scriptions in this part of the world.’’ Id.¶ 308. By ensuring that the ‘‘majority’’ ofall IAP subscribers are offered IE eitheras the default browser or as the only brow-ser, Microsoft’s deals with the IAPs clearlyhave a significant effect in preserving itsmonopoly; they help keep usage of Navi-gator below the critical level necessary forNavigator or any other rival to pose a realthreat to Microsoft’s monopoly. See, e.g.,id. ¶ 143 (Microsoft sought to ‘‘divertenough browser usage from Navigator toneutralize it as a platform.’’); see alsoCarlton, at 670.

Plaintiffs having demonstrated a harmto competition, the burden falls upon Mi-crosoft to defend its exclusive dealingcontracts with IAPs by providing a pro-competitive justification for them. Signif-icantly, Microsoft’s only explanation forits exclusive dealing is that it wants tokeep developers focused upon its APIs—which is to say, it wants to preserve itspower in the operating system market.02/26/01 Ct. Appeals Tr. at 45–47. Thatis not an unlawful end, but neither is it aprocompetitive justification for the specificmeans here in question, namely exclusivedealing contracts with IAPs. According-ly, we affirm the District Court’s decisionholding that Microsoft’s exclusive con-tracts with IAPs are exclusionary devices,in violation of § 2 of the Sherman Act.

4. Dealings with Internet ContentProviders, Independent SoftwareVendors, and Apple Computer

The District Court held that Microsoftengages in exclusionary conduct in itsdealings with ICPs, which develop web-sites; ISVs, which develop software; andApple, which is both an OEM and a soft-ware developer. See Conclusions of Law,

at 42–43 (deals with ICPs, ISVs, and Apple‘‘supplemented Microsoft’s efforts in theOEM and IAP channels’’). The DistrictCourt condemned Microsoft’s deals withICPs and ISVs, stating: ‘‘By grantingICPs and ISVs free licenses to bundle [IE]with their offerings, and by exchangingother valuable inducements for theiragreement to distribute, promote[,] andrely on [IE] rather than Navigator, Micro-soft directly induced developers to focus onits own APIs rather than ones exposed byNavigator.’’ Id. (citing Findings of Fact¶ ¶ 334–35, 340).

[36] With respect to the deals withICPs, the District Court’s findings do notsupport liability. After reviewing the ICPagreements, the District Court specificallystated that ‘‘there is not sufficient evidenceto support a finding that Microsoft’s pro-motional restrictions actually had a sub-stantial, deleterious impact on Navigator’susage share.’’ Findings of Fact ¶ 332. Be-cause plaintiffs failed to demonstrate thatMicrosoft’s deals with the ICPs have asubstantial effect upon competition, theyhave not proved the violation of the Sher-man Act.

[37] As for Microsoft’s ISV agree-ments, however, the District Court did notenter a similar finding of no substantialeffect. The District Court described Mi-crosoft’s deals with ISVs as follows:

In dozens of ‘‘First Wave’’ agreementssigned between the fall of 1997 and thespring of 1998, Microsoft has promisedto give preferential support, in the formof early Windows 98 and Windows NTbetas, other technical information, andthe right to use certain Microsoft sealsof approval, to important ISVs thatagree to certain conditions. One ofthese conditions is that the ISVs useInternet Explorer as the default brows-ing software for any software they de-velop with a hypertext-based user inter-


face. Another condition is that the ISVsuse Microsoft’s ‘‘HTML Help,’’ which isaccessible only with Internet Explorer,to implement their applications’ helpsystems.

Id. ¶ 339. The District Court furtherfound that the effect of these deals is to‘‘ensure [ ] that many of the most popularWeb-centric applications will rely onbrowsing technologies found only in Win-dows,’’ id. ¶ 340, and that Microsoft’s dealswith ISVs therefore ‘‘increase[ ] the likeli-hood that the millions of consumers using[applications designed by ISVs that en-tered into agreements with Microsoft] willuse Internet Explorer rather than Naviga-tor.’’ Id. ¶ 340.

The District Court did not specificallyidentify what share of the market forbrowser distribution the exclusive dealswith the ISVs foreclose. Although theISVs are a relatively small channel forbrowser distribution, they take on greatersignificance because, as discussed above,Microsoft had largely foreclosed the twoprimary channels to its rivals. In thatlight, one can tell from the record that byaffecting the applications used by ‘‘mil-lions’’ of consumers, Microsoft’s exclusivedeals with the ISVs had a substantial ef-fect in further foreclosing rival browsersfrom the market. (Data introduced byMicrosoft, see Direct Testimony of Camer-on Myhrvold ¶ 84, reprinted in 6 J.A. at3922–23, and subsequently relied upon bythe District Court in its findings, see, e.g.,Findings of Fact ¶ 270, indicate that overthe two-year period 1997–98, when Micro-soft entered into the First Wave agree-ments, there were 40 million new users ofthe internet.) Because, by keeping rivalbrowsers from gaining widespread distri-bution (and potentially attracting the at-tention of developers away from the APIsin Windows), the deals have a substantialeffect in preserving Microsoft’s monopoly,we hold that plaintiffs have made a prima

facie showing that the deals have an anti-competitive effect.

Of course, that Microsoft’s exclusivedeals have the anticompetitive effect ofpreserving Microsoft’s monopoly does not,in itself, make them unlawful. A monopo-list, like a competitive firm, may have aperfectly legitimate reason for wanting anexclusive arrangement with its distribu-tors. Accordingly, Microsoft had an op-portunity to, but did not, present the Dis-trict Court with evidence demonstratingthat the exclusivity provisions have somesuch procompetitive justification. SeeConclusions of Law, at 43 (citing Findingsof Fact ¶ ¶ 339–40) (‘‘With respect to theISV agreements, Microsoft has put for-ward no procompetitive business endswhatsoever to justify their exclusionaryterms.’’). On appeal Microsoft likewisedoes not claim that the exclusivity re-quired by the deals serves any legitimatepurpose; instead, it states only that itsISV agreements reflect an attempt ‘‘topersuade ISVs to utilize Internet-relatedsystem services in Windows rather thanNavigator.’’ Appellant’s Opening Br. at114. As we explained before, however,keeping developers focused upon Win-dows—that is, preserving the Windowsmonopoly—is a competitively neutral goal.Microsoft having offered no procompetitivejustification for its exclusive dealing ar-rangements with the ISVs, we hold thatthose arrangements violate § 2 of theSherman Act.

[38] Finally, the District Court heldthat Microsoft’s dealings with Apple violat-ed the Sherman Act. See Conclusions ofLaw, at 42–43. Apple is vertically inte-grated: it makes both software (includingan operating system, Mac OS), and hard-ware (the Macintosh line of computers).Microsoft primarily makes software, in-cluding, in addition to its operating system,


a number of popular applications. One,called ‘‘Office,’’ is a suite of business pro-ductivity applications that Microsoft hasported to Mac OS. The District Courtfound that ‘‘ninety percent of Mac OSusers running a suite of office productivityapplications [use] Microsoft’s Mac Office.’’Findings of Fact ¶ 344. Further, the Dis-trict Court found that:

In 1997, Apple’s business was in steepdecline, and many doubted that the com-pany would survive much longerTTTT

[M]any ISVs questioned the wisdom ofcontinuing to spend time and money de-veloping applications for the Mac OS.Had Microsoft announced in the midst ofthis atmosphere that it was ceasing todevelop new versions of Mac Office, agreat number of ISVs, customers, devel-opers, and investors would have inter-preted the announcement as Apple’sdeath notice.

Id. ¶ 344. Microsoft recognized the impor-tance to Apple of its continued support ofMac Office. See id. ¶ 347 (quoting internalMicrosoft e-mail) (‘‘[We] need a way topush these guys[, i.e., Apple] and [threat-ening to cancel Mac Office] is the only onethat seems to make them move.’’); see alsoid. (‘‘[Microsoft Chairman Bill] Gatesasked whether Microsoft could concealfrom Apple in the coming month the factthat Microsoft was almost finished devel-oping Mac Office 97.’’); id. at ¶ 354 (‘‘Ithink TTT Apple should be using [IE] ev-erywhere and if they don’t do it, then wecan use Office as a club.’’).

In June 1997 Microsoft Chairman BillGates determined that the company’s ne-gotiations with Apple ‘‘ ‘have not been go-ing well at allTTTT Apple let us down onthe browser by making Netscape the stan-dard install.’ Gates then reported that hehad already called Apple’s CEO TTT to ask‘how we should announce the cancellationof Mac OfficeTTTT’ ’’ Id. at ¶ 349. The

District Court further found that, within amonth of Gates’ call, Apple and Microsofthad reached an agreement pursuant towhich

Microsoft’s primary obligation is to con-tinue releasing up-to-date versions ofMac Office for at least five yearsTTTT

[and] Apple has agreed TTT to ‘‘bundlethe most current version of [IE] TTT

with [Mac OS]’’TTT [and to] ‘‘make [IE]the default [browser]’’TTTT Navigator isnot installed on the computer hard driveduring the default installation, which isthe type of installation most users electto employTTTT [The] Agreement fur-ther provides that TTT Apple may notposition icons for nonMicrosoft browsingsoftware on the desktop of new Macin-tosh PC systems or Mac OS upgrades.

Id. ¶ ¶ 350–52. The agreement also pro-hibits Apple from encouraging users tosubstitute another browser for IE, andstates that Apple will ‘‘encourage its em-ployees to use [IE].’’ Id. ¶ 352.

This exclusive deal between Microsoftand Apple has a substantial effect uponthe distribution of rival browsers. If abrowser developer ports its product to asecond operating system, such as theMac OS, it can continue to display a com-mon set of APIs. Thus, usage share, notthe underlying operating system, is theprimary determinant of the platform chal-lenge a browser may pose. Pre-installa-tion of a browser (which can be accom-plished either by including the browserwith the operating system or by theOEM installing the browser) is one of thetwo most important methods of browserdistribution, and Apple had a not insignif-icant share of worldwide sales of operat-ing systems. See id. ¶ 35 (Microsoft has95% of the market not counting Appleand ‘‘well above’’ 80% with Apple includ-ed in the relevant market). Because Mi-crosoft’s exclusive contract with Apple


has a substantial effect in restricting dis-tribution of rival browsers, and because(as we have described several timesabove) reducing usage share of rivalbrowsers serves to protect Microsoft’smonopoly, its deal with Apple must beregarded as anticompetitive. See Conclu-sions of Law, at 42 (citing Findings ofFact ¶ 356) (‘‘By extracting from Appleterms that significantly diminished theusage of Navigator on the Mac OS, Mi-crosoft helped to ensure that developerswould not view Navigator as truly cross-platform middleware.’’).

Microsoft offers no procompetitive justi-fication for the exclusive dealing arrange-ment. It makes only the irrelevant claimthat the IE-for-Mac Office deal is part of amultifaceted set of agreements betweenitself and Apple, see Appellant’s OpeningBr. at 61 (‘‘Apple’s ‘browsing software’ ob-ligation was [not] the quid pro quo forMicrosoft’s Mac Office obligation[;] TTT allof the various obligations TTT were part ofone ‘overall agreement’ between the twocompanies.’’); that does not mean it hasany procompetitive justification. Accord-ingly, we hold that the exclusive deal withApple is exclusionary, in violation of § 2 ofthe Sherman Act.

5. Java

Java, a set of technologies developed bySun Microsystems, is another type of mid-dleware posing a potential threat to Win-dows’ position as the ubiquitous platformfor software development. Findings ofFact ¶ 28. The Java technologies include:(1) a programming language; (2) a set ofprograms written in that language, calledthe ‘‘Java class libraries,’’ which exposeAPIs; (3) a compiler, which translatescode written by a developer into ‘‘bytec-ode’’; and (4) a Java Virtual Machine(‘‘JVM’’), which translates bytecode intoinstructions to the operating system. Id.¶ 73. Programs calling upon the Java

APIs will run on any machine with a ‘‘Javaruntime environment,’’ that is, Java classlibraries and a JVM. Id. ¶ ¶ 73, 74.

In May 1995 Netscape agreed with Sunto distribute a copy of the Java runtimeenvironment with every copy of Navigator,and ‘‘Navigator quickly became the princi-pal vehicle by which Sun placed copies ofits Java runtime environment on the PCsystems of Windows users.’’ Id. ¶ 76. Mi-crosoft, too, agreed to promote the Javatechnologies—or so it seemed. For at thesame time, Microsoft took steps ‘‘to max-imize the difficulty with which applicationswritten in Java could be ported from Win-dows to other platforms, and vice versa.’’Conclusions of Law, at 43. Specifically,the District Court found that Microsofttook four steps to exclude Java from devel-oping as a viable cross-platform threat: (a)designing a JVM incompatible with theone developed by Sun; (b) entering intocontracts, the so-called ‘‘First WaveAgreements,’’ requiring major ISVs topromote Microsoft’s JVM exclusively; (c)deceiving Java developers about the Win-dows-specific nature of the tools it distrib-uted to them; and (d) coercing Intel tostop aiding Sun in improving the Javatechnologies.

a. The incompatible JVM

[39] The District Court held that Mi-crosoft engaged in exclusionary conduct bydeveloping and promoting its own JVM.Conclusions of Law, at 43–44. Sun hadalready developed a JVM for the Windowsoperating system when Microsoft beganwork on its version. The JVM developedby Microsoft allows Java applications torun faster on Windows than does Sun’sJVM, Findings of Fact ¶ 389, but a Javaapplication designed to work with Micro-soft’s JVM does not work with Sun’s JVMand vice versa. Id. ¶ 390. The DistrictCourt found that Microsoft ‘‘made a large


investment of engineering resources to de-velop a high-performance Windows JVM,’’id. ¶ 396, and, ‘‘[b]y bundling its TTT JVMwith every copy of [IE] TTT Microsoft en-dowed its Java runtime environment withthe unique attribute of guaranteed, endur-ing ubiquity across the enormous Windowsinstalled base,’’ id. ¶ 397. As explainedabove, however, a monopolist does not vio-late the antitrust laws simply by develop-ing a product that is incompatible withthose of its rivals. See supra Section II.B.1. In order to violate the antitrust laws,the incompatible product must have ananticompetitive effect that outweighs anyprocompetitive justification for the design.Microsoft’s JVM is not only incompatiblewith Sun’s, it allows Java applications torun faster on Windows than does Sun’sJVM. Microsoft’s faster JVM lured Javadevelopers into using Microsoft’s developertools, and Microsoft offered those toolsdeceptively, as we discuss below. TheJVM, however, does allow applications torun more swiftly and does not itself haveany anticompetitive effect. Therefore, wereverse the District Court’s imposition ofliability for Microsoft’s development andpromotion of its JVM.

b. The First Wave Agreements

[40] The District Court also found thatMicrosoft entered into First Wave Agree-ments with dozens of ISVs to use Micro-soft’s JVM. See Findings of Fact ¶ 401(‘‘[I]n exchange for costly technical sup-port and other blandishments, Microsoftinduced dozens of important ISVs to maketheir Java applications reliant on Win-dows-specific technologies and to refrainfrom distributing to Windows users JVMsthat complied with Sun’s standards.’’).Again, we reject the District Court’s con-demnation of low but non-predatory pric-ing by Microsoft.

To the extent Microsoft’s First WaveAgreements with the ISVs conditioned re-ceipt of Windows technical informationupon the ISVs’ agreement to promote Mi-crosoft’s JVM exclusively, they raise a dif-ferent competitive concern. The DistrictCourt found that, although not literallyexclusive, the deals were exclusive in prac-tice because they required developers tomake Microsoft’s JVM the default in thesoftware they developed. Id. ¶ 401.

While the District Court did not enterprecise findings as to the effect of theFirst Wave Agreements upon the overalldistribution of rival JVMs, the record indi-cates that Microsoft’s deals with the majorISVs had a significant effect upon JVMpromotion. As discussed above, the prod-ucts of First Wave ISVs reached millionsof consumers. Id. ¶ 340. The First WaveISVs included such prominent developersas Rational Software, see GX 970, reprint-ed in 15 J.A. at 9994–10000, ‘‘a worldleader’’ in software development tools, seeDirect Testimony of Michael Devlin ¶ 2,reprinted in 5 J.A. at 3520, and Symantec,see GX 2071, reprinted in 22 J.A. at14960–66 (sealed), which, according to Mi-crosoft itself, is ‘‘the leading supplier ofutilities such as anti-virus software,’’ De-fendant’s Proposed Findings of Fact ¶ 276,reprinted in 3 J.A. at 1689. Moreover,Microsoft’s exclusive deals with the leadingISVs took place against a backdrop offoreclosure: the District Court found that‘‘[w]hen Netscape announced in May 1995[prior to Microsoft’s execution of the FirstWave Agreements] that it would includewith every copy of Navigator a copy of aWindows JVM that complied with Sun’sstandards, it appeared that Sun’s Java im-plementation would achieve the necessaryubiquity on Windows.’’ Findings of Fact¶ 394. As discussed above, however, Mi-crosoft undertook a number of anticompet-itive actions that seriously reduced the dis-tribution of Navigator, and the District


Court found that those actions therebyseriously impeded distribution of Sun’sJVM. Conclusions of Law, at 43–44. Be-cause Microsoft’s agreements foreclosed asubstantial portion of the field for JVMdistribution and because, in so doing, theyprotected Microsoft’s monopoly from amiddleware threat, they are anticompeti-tive.

Microsoft offered no procompetitive jus-tification for the default clause that madethe First Wave Agreements exclusive as apractical matter. See Findings of Fact¶ 401. Because the cumulative effect ofthe deals is anticompetitive and becauseMicrosoft has no procompetitive justifica-tion for them, we hold that the provisionsin the First Wave Agreements requiringuse of Microsoft’s JVM as the default areexclusionary, in violation of the ShermanAct.

c. Deception of Java developers

[41] Microsoft’s ‘‘Java implementation’’included, in addition to a JVM, a set ofsoftware development tools it created toassist ISVs in designing Java applications.The District Court found that, not onlywere these tools incompatible with Sun’scross-platform aspirations for Java—no vi-olation, to be sure—but Microsoft deceivedJava developers regarding the Windows-specific nature of the tools. Microsoft’stools included ‘‘certain ‘keywords’ and‘compiler directives’ that could only be exe-cuted properly by Microsoft’s version ofthe Java runtime environment for Win-dows.’’ Id. ¶ 394; see also Direct Testimo-ny of James Gosling ¶ 58, reprinted in 21J.A. at 13959 (Microsoft added ‘‘program-ming instructions TTT that alter the behav-ior of the code.’’). As a result, even Java‘‘developers who were opting for portabili-ty over performance TTT unwittingly[wrote] Java applications that [ran] only onWindows.’’ Conclusions of Law, at 43.

That is, developers who relied upon Micro-soft’s public commitment to cooperate withSun and who used Microsoft’s tools todevelop what Microsoft led them to believewere cross-platform applications ended upproducing applications that would run onlyon the Windows operating system.

When specifically accused by a PC Weekreporter of fragmenting Java standards soas to prevent cross-platform uses, Micro-soft denied the accusation and indicated itwas only ‘‘adding rich platform support’’ towhat remained a crossplatform implemen-tation. An e-mail message internal to Mi-crosoft, written shortly after the conversa-tion with the reporter, shows otherwise:

[O]k, i just did a followup callTTTT [Thereporter] liked that i kept pointing cus-tomers to w3c standards [ (commonlyobserved internet protocols) ]TTTT [but]he accused us of being schizo with thisvs. our java approach, i said he misun-derstood [—] that [with Java] we aremerely trying to add rich platform sup-port to an interop layerTTTT this playswellTTTT at this point its [sic] not goodto create MORE noise around our win32java classes. instead we should justquietly grow jvv [ (Microsoft’s devel-opment tools) ] share and assume thatpeople will take more advantage of ourclasses without ever realizing they arebuilding win32–only java apps.

GX 1332, reprinted in 22 J.A. at 14922–23.

Finally, other Microsoft documents con-firm that Microsoft intended to deceiveJava developers, and predicted that theeffect of its actions would be to generateWindows-dependent Java applications thattheir developers believed would be cross-platform; these documents also indicatethat Microsoft’s ultimate objective was tothwart Java’s threat to Microsoft’s monop-oly in the market for operating systems.One Microsoft document, for example,states as a strategic goal: ‘‘Kill cross-plat-


form Java by grow[ing] the polluted Javamarket.’’ GX 259, reprinted in 22 J.A. at14514; see also id. (‘‘Cross-platform capa-bility is by far the number one reason forchoosing/using Java.’’) (emphasis in origi-nal).

Microsoft’s conduct related to its Javadeveloper tools served to protect its mo-nopoly of the operating system in a man-ner not attributable either to the superi-ority of the operating system or to theacumen of its makers, and therefore wasanticompetitive. Unsurprisingly, Micro-soft offers no procompetitive explanationfor its campaign to deceive developers.Accordingly, we conclude this conduct isexclusionary, in violation of § 2 of theSherman Act.

d. The threat to Intel

[42] The District Court held that Mi-crosoft also acted unlawfully with respectto Java by using its ‘‘monopoly power toprevent firms such as Intel from aiding inthe creation of cross-platform interfaces.’’Conclusions of Law, at 43. In 1995 Intelwas in the process of developing a highper-formance, Windows-compatible JVM. Mi-crosoft wanted Intel to abandon that effortbecause a fast, cross-platform JVM wouldthreaten Microsoft’s monopoly in the oper-ating system market. At an August 1995meeting, Microsoft’s Gates told Intel thatits ‘‘cooperation with Sun and Netscape todevelop a Java runtime environment TTT

was one of the issues threatening to under-mine cooperation between Intel and Micro-soft.’’ Findings of Fact ¶ 396. Threemonths later, ‘‘Microsoft’s Paul Maritz tolda senior Intel executive that Intel’s [adap-tation of its multimedia software to complywith] Sun’s Java standards was as inimicalto Microsoft as Microsoft’s support fornon-Intel microprocessors would be to In-tel.’’ Id. ¶ 405.

Intel nonetheless continued to undertakeinitiatives related to Java. By 1996 ‘‘Intelhad developed a JVM designed to run wellTTT while complying with Sun’s cross-plat-form standards.’’ Id. ¶ 396. In April ofthat year, Microsoft again urged Intel notto help Sun by distributing Intel’s fast,Suncompliant JVM. Id. And Microsoftthreatened Intel that if it did not stopaiding Sun on the multimedia front, thenMicrosoft would refuse to distribute Inteltechnologies bundled with Windows. Id.¶ 404.

Intel finally capitulated in 1997, afterMicrosoft delivered the coup de grace.

[O]ne of Intel’s competitors, calledAMD, solicited support from Microsoftfor its ‘‘3DX’’ technologyTTTT Micro-soft’s Allchin asked Gates whether Mi-crosoft should support 3DX, despite thefact that Intel would oppose it. Gatesresponded: ‘‘If Intel has a real problemwith us supporting this then they willhave to stop supporting Java Multimediathe way they are. I would gladly giveup supporting this if they would back offfrom their work on JAVA.’’

Id. ¶ 406.

Microsoft’s internal documents and de-position testimony confirm both the anti-competitive effect and intent of its actions.See, e.g., GX 235, reprinted in 22 J.A. at14502 (Microsoft executive, Eric Eng-strom, included among Microsoft’s goalsfor Intel: ‘‘Intel to stop helping Sun createJava Multimedia APIs, especially ones thatrun well TTT on Windows.’’); Deposition ofEric Engstrom at 179 (‘‘We were success-ful [in convincing Intel to stop aiding Sun]for some period of time.’’).

Microsoft does not deny the facts foundby the District Court, nor does it offer anyprocompetitive justification for pressuringIntel not to support cross-platform Java.Microsoft lamely characterizes its threat toIntel as ‘‘advice.’’ The District Court,


however, found that Microsoft’s ‘‘advice’’ toIntel to stop aiding cross-platform Javawas backed by the threat of retaliation,and this conclusion is supported by theevidence cited above. Therefore we affirmthe conclusion that Microsoft’s threats toIntel were exclusionary, in violation of § 2of the Sherman Act.

6. Course of Conduct

[43] The District Court held that,apart from Microsoft’s specific acts, Micro-soft was liable under § 2 based upon itsgeneral ‘‘course of conduct.’’ In reachingthis conclusion the court relied upon Conti-nental Ore Co. v. Union Carbide & CarbonCorp., 370 U.S. 690, 699, 82 S.Ct. 1404, 8L.Ed.2d 777 (1962), where the SupremeCourt stated, ‘‘[i]n [Sherman Act cases],plaintiffs should be given the full benefit oftheir proof without tightly compartmental-izing the various factual components andwiping the slate clean after scrutiny ofeach.’’

Microsoft points out that ContinentalOre and the other cases cited by plaintiffsin support of ‘‘course of conduct’’ liabilityall involve conspiracies among multiplefirms, not the conduct of a single firm; inthat setting the ‘‘course of conduct’’ is theconspiracy itself, for which all the partici-pants may be held liable. See Appellant’sOpening Br. at 112–13. Plaintiffs respondthat, as a policy matter, a monopolist’sunilateral ‘‘campaign of [acts intended toexclude a rival] that in the aggregate hasthe requisite impact’’ warrants liabilityeven if the acts viewed individually wouldbe lawful for want of a significant effectupon competition. Appellees’ Br. at 82–83.

We need not pass upon plaintiffs’ argu-ment, however, because the District Courtdid not point to any series of acts, each ofwhich harms competition only slightly butthe cumulative effect of which is significantenough to form an independent basis forliability. The ‘‘course of conduct’’ section

of the District Court’s opinion contains,with one exception, only broad, summariz-ing conclusions. See, e.g., Conclusions ofLaw, at 44 (‘‘Microsoft placed an oppres-sive thumb on the scale of competitivefortuneTTTT’’). The only specific acts towhich the court refers are Microsoft’s ex-penditures in promoting its browser, seeid. (‘‘Microsoft has expended wealth andforesworn opportunities to realizemoreTTTT’’), which we have explained arenot in themselves unlawful. Because theDistrict Court identifies no other specificacts as a basis for ‘‘course of conduct’’liability, we reverse its conclusion that Mi-crosoft’s course of conduct separately vio-lates § 2 of the Sherman Act.

C. Causation

[44] As a final parry, Microsoft urgesthis court to reverse on the monopolymaintenance claim, because plaintiffs nev-er established a causal link between Mi-crosoft’s anticompetitive conduct, in partic-ular its foreclosure of Netscape’s andJava’s distribution channels, and the main-tenance of Microsoft’s operating systemmonopoly. See Findings of Fact ¶ 411(‘‘There is insufficient evidence to findthat, absent Microsoft’s actions, Navigatorand Java already would have ignited genu-ine competition in the market for Intel-compatible PC operating systems.’’). Thisis the flip side of Microsoft’s earlier argu-ment that the District Court should haveincluded middleware in the relevant mar-ket. According to Microsoft, the DistrictCourt cannot simultaneously find that mid-dleware is not a reasonable substitute andthat Microsoft’s exclusionary conduct con-tributed to the maintenance of monopolypower in the operating system market.Microsoft claims that the first finding de-pended on the court’s view that middle-ware does not pose a serious threat toWindows, see supra Section II.A, while the


second finding required the court to findthat Navigator and Java would have devel-oped into serious enough cross-platformthreats to erode the applications barrier toentry. We disagree.

Microsoft points to no case, and we canfind none, standing for the propositionthat, as to § 2 liability in an equitableenforcement action, plaintiffs must presentdirect proof that a defendant’s continuedmonopoly power is precisely attributableto its anticompetitive conduct. As its loneauthority, Microsoft cites the followingpassage from Professor Areeda’s antitrusttreatise: ‘‘The plaintiff has the burden ofpleading, introducing evidence, and pre-sumably proving by a preponderance ofthe evidence that reprehensible behaviorhas contributed significantly to the TTT

maintenance of the monopoly.’’ 3 PHILLIP

E. AREEDA & HERBERT HOVENKAMP, ANTI-

TRUST LAW ¶ 650c, at 69 (1996) (emphasisadded).

But, with respect to actions seeking in-junctive relief, the authors of that treatisealso recognize the need for courts to infer‘‘causation’’ from the fact that a defendanthas engaged in anticompetitive conductthat ‘‘reasonably appear[s] capable ofmaking a significant contribution to TTT

maintaining monopoly power.’’ Id. ¶ 651c,at 78; see also Morgan v. Ponder, 892F.2d 1355, 1363 (8th Cir.1989); BarryWright, 724 F.2d at 230. To require that§ 2 liability turn on a plaintiff’s ability orinability to reconstruct the hypotheticalmarketplace absent a defendant’s anticom-petitive conduct would only encourage mo-nopolists to take more and earlier anti-competitive action.

We may infer causation when exclu-sionary conduct is aimed at producers ofnascent competitive technologies as wellas when it is aimed at producers of es-tablished substitutes. Admittedly, in theformer case there is added uncertainty,

inasmuch as nascent threats are merelypotential substitutes. But the underlyingproof problem is the same—neither plain-tiffs nor the court can confidently recon-struct a product’s hypothetical technologi-cal development in a world absent thedefendant’s exclusionary conduct. Tosome degree, ‘‘the defendant is made tosuffer the uncertain consequences of itsown undesirable conduct.’’ 3 AREEDA &HOVENKAMP, ANTITRUST LAW ¶ 651c, at 78.

Given this rather edentulous test forcausation, the question in this case is notwhether Java or Navigator would actuallyhave developed into viable platform substi-tutes, but (1) whether as a general matterthe exclusion of nascent threats is the typeof conduct that is reasonably capable ofcontributing significantly to a defendant’scontinued monopoly power and (2) whetherJava and Navigator reasonably constitutednascent threats at the time Microsoft en-gaged in the anticompetitive conduct atissue. As to the first, suffice it to say thatit would be inimical to the purpose of theSherman Act to allow monopolists freereign to squash nascent, albeit unproven,competitors at will—particularly in indus-tries marked by rapid technological ad-vance and frequent paradigm shifts.Findings of Fact ¶ ¶ 59–60. As to thesecond, the District Court made amplefindings that both Navigator and Javashowed potential as middleware platformthreats. Findings of Fact ¶ ¶ 68–77.Counsel for Microsoft admitted as much atoral argument. 02/26/01 Ct. Appeals Tr.at 27 (‘‘There are no constraints on output.Marginal costs are essentially zero. Andthere are to some extent network effects.So a company like Netscape founded in1994 can be by the middle of 1995 clearly apotentially lethal competitor to Windowsbecause it can supplant its position in themarket because of the characteristics ofthese markets.’’).


Microsoft’s concerns over causationhave more purchase in connection with theappropriate remedy issue, i.e., whetherthe court should impose a structural reme-dy or merely enjoin the offensive conductat issue. As we point out later in thisopinion, divestiture is a remedy that is im-posed only with great caution, in part be-cause its long-term efficacy is rarely cer-tain. See infra Section V.E. Absentsome measure of confidence that there hasbeen an actual loss to competition thatneeds to be restored, wisdom counselsagainst adopting radical structural relief.See 3 AREEDA & HOVENKAMP, ANTITRUST

LAW ¶ 653b, at 91–92 (‘‘[M]ore extensiveequitable relief, particularly remedies suchas divestiture designed to eliminate themonopoly altogether, raise more seriousquestions and require a clearer indicationof a significant causal connection betweenthe conduct and creation or maintenanceof the market power.’’). But these queriesgo to questions of remedy, not liability.In short, causation affords Microsoft nodefense to liability for its unlawful actionsundertaken to maintain its monopoly inthe operating system market.

III. ATTEMPTED MONOPOLIZATION

[45] Microsoft further challenges theDistrict Court’s determination of liabilityfor ‘‘attempt[ing] to monopolize TTT anypart of the trade or commerce among theseveral States.’’ 15 U.S.C. § 2 (1997). Toestablish a § 2 violation for attempted mo-nopolization, ‘‘a plaintiff must prove (1)that the defendant has engaged in predato-ry or anticompetitive conduct with (2) aspecific intent to monopolize and (3) a dan-gerous probability of achieving monopolypower.’’ Spectrum Sports, Inc. v. McQuil-lan, 506 U.S. 447, 456, 113 S.Ct. 884, 122L.Ed.2d 247 (1993); see also Times–Picay-une Pub. Co. v. United States, 345 U.S.594, 626, 73 S.Ct. 872, 97 L.Ed. 1277(1953); Lorain Journal Co. v. United

States, 342 U.S. 143, 153–55, 72 S.Ct. 181,96 L.Ed. 162 (1951). Because a deficiencyon any one of the three will defeat plain-tiffs’ claim, we look no further than plain-tiffs’ failure to prove a dangerous probabil-ity of achieving monopoly power in theputative browser market.

[46] The determination whether a dan-gerous probability of success exists is aparticularly fact-intensive inquiry. Be-cause the Sherman Act does not identifythe activities that constitute the offense ofattempted monopolization, the court ‘‘mustexamine the facts of each case, mindfulthat the determination of what constitutesan attempt, as Justice Holmes explained,‘is a question of proximity and degree.’ ’’United States v. Am. Airlines, Inc., 743F.2d 1114, 1118 (5th Cir.1984) (quotingSwift & Co. v. United States, 196 U.S.375, 402, 25 S.Ct. 276, 49 L.Ed. 518(1905)). The District Court determinedthat ‘‘[t]he evidence supports the conclu-sion that Microsoft’s actions did pose sucha danger.’’ Conclusions of Law, at 45.Specifically, the District Court concludedthat ‘‘Netscape’s assent to Microsoft’smarket division proposal would have, in-stanter, resulted in Microsoft’s attainmentof monopoly power in a second market,’’and that ‘‘the proposal itself created adangerous probability of that result.’’Conclusions of Law, at 46 (citation omit-ted). The District Court further conclud-ed that ‘‘the predatory course of conductMicrosoft has pursued since June of 1995has revived the dangerous probability thatMicrosoft will attain monopoly power in asecond market.’’ Id.

At the outset we note a pervasive flawin the District Court’s and plaintiffs’ dis-cussion of attempted monopolization.Simply put, plaintiffs have made the sameargument under two different headings—monopoly maintenance and attempted mo-


nopolization. They have relied upon Mi-crosoft’s § 2 liability for monopolization ofthe operating system market as a pre-sumptive indicator of attempted monopoli-zation of an entirely different market.The District Court implicitly accepted thisapproach: It agreed with plaintiffs thatthe events that formed the basis for the§ 2 monopolization claim ‘‘warrant[ed] ad-ditional liability as an illegal attempt toamass monopoly power in ‘the browsermarket.’ ’’ Id. at 45 (emphasis added).Thus, plaintiffs and the District Courtfailed to recognize the need for an analysiswholly independent of the conclusions andfindings on monopoly maintenance.

To establish a dangerous probability ofsuccess, plaintiffs must as a threshold mat-ter show that the browser market can bemonopolized, i.e., that a hypothetical mo-nopolist in that market could enjoy marketpower. This, in turn, requires plaintiffs(1) to define the relevant market and (2) todemonstrate that substantial barriers toentry protect that market. Because plain-tiffs have not carried their burden on ei-ther prong, we reverse without remand.

A. Relevant Market

[47, 48] A court’s evaluation of an at-tempted monopolization claim must includea definition of the relevant market. SeeSpectrum Sports, 506 U.S. at 455–56, 113S.Ct. 884. Such a definition establishes acontext for evaluating the defendant’s ac-tions as well as for measuring whether thechallenged conduct presented a dangerousprobability of monopolization. See id.The District Court omitted this element ofthe Spectrum Sports inquiry.

Defining a market for an attempted mo-nopolization claim involves the same stepsas defining a market for a monopoly main-tenance claim, namely a detailed descrip-tion of the purpose of a browser—whatfunctions may be included and what are

not—and an examination of the substitutesthat are part of the market and those thatare not. See also supra Section II.A. TheDistrict Court never engaged in such ananalysis nor entered detailed findings de-fining what a browser is or what productsmight constitute substitutes. In the Find-ings of Fact, the District Court (in a sec-tion on whether IE and Windows are sepa-rate products) stated only that ‘‘a Webbrowser provides the ability for the enduser to select, retrieve, and perceive re-sources on the Web.’’ Findings of Fact¶ 150. Furthermore, in discussing at-tempted monopolization in its Conclusionsof Law, the District Court failed to demon-strate analytical rigor when it employedvarying and imprecise references to the‘‘market for browsing technology for Win-dows,’’ ‘‘the browser market,’’ and ‘‘plat-form-level browsing software.’’ Conclu-sions of Law, at 45.

[49] Because the determination of arelevant market is a factual question to beresolved by the District Court, see, e.g., AllCare Nursing Serv., Inc. v. High TechStaffing Servs., Inc., 135 F.3d 740, 749(11th Cir.1998); Tunis Bros. Co., Inc. v.Ford Motor Co., 952 F.2d 715, 722–23 (3dCir.1991); Westman Comm’n Co. v. Ho-bart Int’l, Inc., 796 F.2d 1216, 1220 (10thCir.1986), we would normally remand thecase so that the District Court could for-mulate an appropriate definition. SeePullman–Standard v. Swint, 456 U.S. 273,291–92 & n. 22, 102 S.Ct. 1781, 72 L.Ed.2d66 (1982); Janini v. Kuwait Univ., 43F.3d 1534, 1537 (D.C.Cir.1995); Palmer v.Shultz, 815 F.2d 84, 103 (D.C.Cir.1987). Aremand on market definition is unneces-sary, however, because the District Court’simprecision is directly traceable to plain-tiffs’ failure to articulate and identify evi-dence before the District Court as to (1)what constitutes a browser (i.e., what arethe technological components of or func-


tionalities provided by a browser) and (2)why certain other products are not reason-able substitutes (e.g., browser shells orviewers for individual internet extensions,such as Real Audio Player or Adobe Acro-bat Reader). See Plaintiffs’ Joint Pro-posed Findings of Fact, at 817–19, reprint-ed in 2 J.A. at 1480–82; Plaintiffs’ JointProposed Conclusions of Law § IV (No.98–1232); see also Lee v. Interstate Fire &Cas. Co., 86 F.3d 101, 105 (7th Cir.1996)(stating that remand for development of afactual record is inappropriate whereplaintiff failed to meet burden of persua-sion and never suggested that additionalevidence was necessary). Indeed, whenplaintiffs in their Proposed Findings ofFact attempted to define a relevant mar-ket for the attempt claim, they pointedonly to their separate products analysis forthe tying claim. See, e.g., Plaintiffs’ JointProposed Findings of Fact, at 818, reprint-ed in 2 J.A. at 1481. However, the sepa-rate products analysis for tying purposesis not a substitute for the type of marketdefinition that Spectrum Sports requires.See infra Section IV.A.

Plaintiffs’ proposed findings and the Dis-trict Court’s actual findings on attemptedmonopolization pale in comparison to theircounterparts on the monopoly maintenanceclaim. Compare Findings of Fact ¶ 150,and Plaintiffs’ Joint Proposed Findings ofFact, at 817–819, reprinted in 2 J.A. at1480–82, with Findings of Fact ¶ ¶ 18–66,and Plaintiffs’ Joint Proposed Findings ofFact, at 20–31, reprinted in 1 J.A. at 658–69. Furthermore, in their brief and atoral argument before this court, plaintiffsdid nothing to clarify or ameliorate thisdeficiency. See, e.g., Appellees’ Br. at 93–94.

B. Barriers to Entry

[50, 51] Because a firm cannot possessmonopoly power in a market unless that

market is also protected by significantbarriers to entry, see supra Section II.A,it follows that a firm cannot threaten toachieve monopoly power in a market un-less that market is, or will be, similarlyprotected. See Spectrum Sports, 506 U.S.at 456, 113 S.Ct. 884 (‘‘In order to deter-mine whether there is a dangerous proba-bility of monopolization, courts have foundit necessary to consider TTT the defen-dant’s ability to lessen or destroy competi-tion in that market.’’) (citing cases).Plaintiffs have the burden of establishingbarriers to entry into a properly definedrelevant market. See 2A PHILLIP E. AREE-

DA ET AL., ANTITRUST LAW ¶ 420b, at 57–59(1995); 3A PHILLIP E. AREEDA & HERBERT

HOVENKAMP, ANTITRUST LAW ¶ 807g, at 361–62 (1996); see also Neumann v. Rein-forced Earth Co., 786 F.2d 424, 429(D.C.Cir.1986). Plaintiffs must not onlyshow that barriers to entry protect theproperly defined browser market, but thatthose barriers are ‘‘significant.’’ See Unit-ed States v. Baker Hughes Inc., 908 F.2d981, 987 (D.C.Cir.1990). Whether thereare significant barriers to entry cannot, ofcourse, be answered absent an appropriatemarket definition; thus, plaintiffs’ failureon that score alone is dispositive. Buteven were we to assume a properly de-fined market, for example browsers con-sisting of a graphical interface plus inter-net protocols, plaintiffs nonetheless failedto carry their burden on barriers to entry.

Contrary to plaintiffs’ contention on ap-peal, see Appellees’ Br. at 91–93, none ofthe District Court’s statements constitutesa finding of barriers to entry into the webbrowser market. Finding of Fact 89states:

At the time Microsoft presented its pro-posal, Navigator was the only browserproduct with a significant share of themarket and thus the only one with thepotential to weaken the applications bar-rier to entry. Thus, had it convinced


Netscape to accept its offer of a ‘‘specialrelationship,’’ Microsoft quickly wouldhave gained such control over the exten-sions and standards that networkcentricapplications (including Web sites) em-ploy as to make it all but impossible forany future browser rival to lure appreci-able developer interest away from Mi-crosoft’s platform.

This finding is far too speculative toestablish that competing browsers wouldbe unable to enter the market, or thatMicrosoft would have the power to raisethe price of its browser above, or reducethe quality of its browser below, the com-petitive level. Moreover, it is ambiguousinsofar as it appears to focus on Micro-soft’s response to the perceived platformthreat rather than the browser market.Finding of Fact 144, on which plaintiffsalso rely, is part of the District Court’sdiscussion of Microsoft’s alleged anticom-petitive actions to eliminate the platformthreat posed by Netscape Navigator. Thisfinding simply describes Microsoft’s reli-ance on studies indicating consumers’ re-luctance to switch browsers, a reluctancenot shown to be any more than that whichstops consumers from switching brands ofcereal. Absent more extensive and defini-tive factual findings, the District Court’slegal conclusions about entry barriersamount to nothing more than speculation.

In contrast to their minimal effort onmarket definition, plaintiffs did at leastoffer proposed findings of fact suggestingthat the possibility of network effects couldpotentially create barriers to entry into thebrowser market. See Plaintiffs’ Joint Pro-posed Findings of Fact, at 822–23, 825–27,reprinted in 2 J.A. at 1485–86, 1488–90.The District Court did not adopt thoseproposed findings. See Findings of Fact¶ 89. However, the District Court did ac-knowledge the possibility of a different

kind of entry barrier in its Conclusions ofLaw:

In the time it would have taken an aspir-ing entrant to launch a serious effort tocompete against Internet Explorer, Mi-crosoft could have erected the sametype of barrier that protects its existingmonopoly power by adding proprietaryextensions to the browsing software un-der its control and by extracting com-mitments from OEMs, IAPs and otherssimilar to the ones discussed in [themonopoly maintenance section].

Conclusions of Law, at 46 (emphasis add-ed).

Giving plaintiffs and the District Courtthe benefit of the doubt, we might remandif the possible existence of entry barriersresulting from the possible creation andexploitation of network effects in thebrowser market were the only concern.That is not enough to carry the day, how-ever, because the District Court did notmake two key findings: (1) that networkeffects were a necessary or even probable,rather than merely possible, consequenceof high market share in the browser mar-ket and (2) that a barrier to entry result-ing from network effects would be ‘‘signifi-cant’’ enough to confer monopoly power.Again, these deficiencies are in large parttraceable to plaintiffs’ own failings. As tothe first point, the District Court’s use ofthe phrase ‘‘could have’’ reflects the sameuncertainty articulated in testimony citedin plaintiffs’ proposed findings. See Plain-tiffs’ Joint Proposed Findings of Fact, at822 (citing testimony of Frederick War-ren–Boulton), at 826 (citing testimony ofFranklin Fisher), reprinted in 2 J.A. at1485, 1489. As to the second point, thecited testimony in plaintiffs’ proposed find-ings offers little more than conclusorystatements. See id. at 822–27, reprintedin 2 J.A. at 1485–90. The proffered testi-mony contains no evidence regarding the


cost of ‘‘porting’’ websites to differentbrowsers or the potentially different eco-nomic incentives facing ICPs, as opposedto ISVs, in their decision to incur costs todo so. Simply invoking the phrase ‘‘net-work effects’’ without pointing to more evi-dence does not suffice to carry plaintiffs’burden in this respect.

Any doubt that we may have had re-garding remand instead of outright rever-sal on the barriers to entry question wasdispelled by plaintiffs’ arguments on at-tempted monopolization before this court.Not only did plaintiffs fail to articulate awebsite barrier to entry theory in eithertheir brief or at oral argument, they failedto point the court to evidence in the recordthat would support a finding that Micro-soft would likely erect significant barriersto entry upon acquisition of a dominantmarket share.

Plaintiffs did not devote the same re-sources to the attempted monopolizationclaim as they did to the monopoly mainte-nance claim. But both claims require evi-dentiary and theoretical rigor. Becauseplaintiffs failed to make their case on at-tempted monopolization both in the Dis-trict Court and before this court, there isno reason to give them a second chance toflesh out a claim that should have beenfleshed out the first time around. Accord-ingly, we reverse the District Court’s de-termination of § 2 liability for attemptedmonopolization.

IV. TYING

[52, 53] Microsoft also contests theDistrict Court’s determination of liabilityunder § 1 of the Sherman Act. The Dis-trict Court concluded that Microsoft’s con-tractual and technological bundling of theIE web browser (the ‘‘tied’’ product) withits Windows operating system (‘‘OS’’) (the‘‘tying’’ product) resulted in a tying ar-rangement that was per se unlawful. Con-

clusions of Law, at 47–51. We hold thatthe rule of reason, rather than per seanalysis, should govern the legality of ty-ing arrangements involving platform soft-ware products. The Supreme Court haswarned that ‘‘ ‘[i]t is only after considera-ble experience with certain business rela-tionships that courts classify them as perse violationsTTTT’ ’’ Broad. Music, Inc. v.CBS, 441 U.S. 1, 9, 99 S.Ct. 1551, 60L.Ed.2d 1 (1979) (quoting United States v.Topco Assocs., 405 U.S. 596, 607–08, 92S.Ct. 1126, 31 L.Ed.2d 515 (1972)). Whileevery ‘‘business relationship’’ will in somesense have unique features, some repre-sent entire, novel categories of dealings.As we shall explain, the arrangement be-fore us is an example of the latter, offeringthe first up-close look at the technologicalintegration of added functionality into soft-ware that serves as a platform for third-party applications. There being no closeparallel in prior antitrust cases, simplisticapplication of per se tying rules carries aserious risk of harm. Accordingly, we va-cate the District Court’s finding of a per setying violation and remand the case.Plaintiffs may on remand pursue their ty-ing claim under the rule of reason.

The facts underlying the tying allegationsubstantially overlap with those set forthin Section II.B in connection with the § 2monopoly maintenance claim. The keyDistrict Court findings are that (1) Micro-soft required licensees of Windows 95 and98 also to license IE as a bundle at a singleprice, Findings of Fact ¶ ¶ 137, 155, 158;(2) Microsoft refused to allow OEMs touninstall or remove IE from the Windowsdesktop, id. ¶ ¶ 158, 203, 213; (3) Microsoftdesigned Windows 98 in a way that with-held from consumers the ability to removeIE by use of the Add/Remove Programsutility, id. ¶ 170; cf. id. ¶ 165 (stating thatIE was subject to Add/Remove Programsutility in Windows 95); and (4) Microsoft


designed Windows 98 to override theuser’s choice of default web browser incertain circumstances, id. ¶ ¶ 171, 172.The court found that these acts constituteda per se tying violation. Conclusions ofLaw, at 47–51. Although the DistrictCourt also found that Microsoft commin-gled operating system-only and browser-only routines in the same library files,Findings of Fact ¶ ¶ 161, 164, it did notinclude this as a basis for tying liabilitydespite plaintiffs’ request that it do so,Plaintiffs’ Proposed Findings of Fact,¶ ¶ 131–32, reprinted in 2 J.A. at 941–47.

[54] There are four elements to a perse tying violation: (1) the tying and tiedgoods are two separate products; (2) thedefendant has market power in the tyingproduct market; (3) the defendant affordsconsumers no choice but to purchase thetied product from it; and (4) the tyingarrangement forecloses a substantial vol-ume of commerce. See Eastman KodakCo. v. Image Tech. Servs., Inc., 504 U.S.451, 461–62, 112 S.Ct. 2072, 119 L.Ed.2d265 (1992); Jefferson Parish Hosp. Dist.No. 2 v. Hyde, 466 U.S. 2, 12–18, 104 S.Ct.1551, 80 L.Ed.2d 2 (1984).

Microsoft does not dispute that it boundWindows and IE in the four ways theDistrict Court cited. Instead it arguesthat Windows (the tying good) and IEbrowsers (the tied good) are not ‘‘separateproducts,’’ Appellant’s Opening Br. at 69–79, and that it did not substantially fore-close competing browsers from the tiedproduct market, id. at 79–83. (Microsoftalso contends that it does not have monop-oly power in the tying product market, id.at 84–96, but, for reasons given in SectionII.A, we uphold the District Court’s find-ing to the contrary.)

We first address the separate-productsinquiry, a source of much argument be-tween the parties and of confusion in the

cases. Our purpose is to highlight thepoor fit between the separate-products testand the facts of this case. We then offerfurther reasons for carving an exception tothe per se rule when the tying product isplatform software. In the final section wediscuss the District Court’s inquiry ifplaintiffs pursue a rule of reason claim onremand.

A. Separate-Products Inquiry Under thePer Se Test

The requirement that a practice involvetwo separate products before being con-demned as an illegal tie started as a purelylinguistic requirement: unless productsare separate, one cannot be ‘‘tied’’ to theother. Indeed, the nature of the productsinvolved in early tying cases—intuitivelydistinct items such as a movie projectorand a film, Motion Picture Patents Co. v.Universal Film Mfg. Co., 243 U.S. 502, 37S.Ct. 416, 61 L.Ed. 871 (1917)—led courtseither to disregard the separate-productsquestion, see, e.g., United Shoe Mach.Corp. v. United States, 258 U.S. 451, 42S.Ct. 363, 66 L.Ed. 708 (1922), or to dis-cuss it only in passing, see, e.g., MotionPicture Patents, 243 U.S. at 508, 512, 518,37 S.Ct. 416. It was not until Times-Picayune Publishing Co. v. United States,345 U.S. 594, 73 S.Ct. 872, 97 L.Ed. 1277(1953), that the separate-products issue be-came a distinct element of the test for anillegal tie. Id. at 614, 73 S.Ct. 872. Eventhat case engaged in a rather cursory in-quiry into whether ads sold in the morningedition of a paper were a separate productfrom ads sold in the evening edition.

The first case to give content to theseparate-products test was Jefferson Par-ish, 466 U.S. 2, 104 S.Ct. 1551, 80 L.Ed.2d2. That case addressed a tying arrange-ment in which a hospital conditioned surgi-cal care at its facility on the purchase ofanesthesiological services from an affiliat-


ed medical group. The facts were a chal-lenge for casual separate-products analysisbecause the tied service—anesthesia—wasneither intuitively distinct from nor intui-tively contained within the tying service—surgical care. A further complication wasthat, soon after the Court enunciated theper se rule for tying liability in Interna-tional Salt Co. v. United States, 332 U.S.392, 396, 68 S.Ct. 12, 92 L.Ed. 20 (1947),and Northern Pacific Railway Co. v. Unit-ed States, 356 U.S. 1, 5–7, 78 S.Ct. 514, 2L.Ed.2d 545 (1958), new economic researchbegan to cast doubt on the assumption,voiced by the Court when it establishedthe rule, that ‘‘ ‘tying agreements servehardly any purpose beyond the suppres-sion of competition,’ ’’ id. at 6, 78 S.Ct. 514(quoting Standard Oil of Cal. v. UnitedStates, 337 U.S. 293, 305–06, 69 S.Ct. 1051,93 L.Ed. 1371 (1949)); see also JeffersonParish, 466 U.S. at 15 n. 23, 104 S.Ct. 1551(citing materials); Fortner Enters. v. U.S.Steel Corp., 394 U.S. 495, 524–25, 89 S.Ct.1252, 22 L.Ed.2d 495 (1969) (Fortas, J.,dissenting) (‘‘Fortner I’’).

The Jefferson Parish Court resolved thematter in two steps. First, it clarified that‘‘the answer to the question whether oneor two products are involved’’ does notturn ‘‘on the functional relation betweenthemTTTT’’ Jefferson Parish, 466 U.S. at19, 104 S.Ct. 1551; see also id. at 19 n. 30,104 S.Ct. 1551. In other words, the merefact that two items are complements, that‘‘one TTT is useless without the other,’’ id.,does not make them a single ‘‘product’’ forpurposes of tying law. Accord EastmanKodak, 504 U.S. at 463, 112 S.Ct. 2072.Second, reasoning that the ‘‘definitionalquestion [whether two distinguishableproducts are involved] depends on whetherthe arrangement may have the type ofcompetitive consequences addressed bythe rule [against tying],’’ Jefferson Parish,466 U.S. at 21, 104 S.Ct. 1551, the Courtdecreed that ‘‘no tying arrangement can

exist unless there is a sufficient demandfor the purchase of anesthesiological ser-vices separate from hospital services toidentify a distinct product market in whichit is efficient to offer anesthesiological ser-vices separately from hospital service,’’ id.at 21–22, 104 S.Ct. 1551 (emphasis added);accord Eastman Kodak, 504 U.S. at 462,112 S.Ct. 2072.

The Court proceeded to examine directand indirect evidence of consumer demandfor the tied product separate from thetying product. Direct evidence addressesthe question whether, when given a choice,consumers purchase the tied good fromthe tying good maker, or from other firms.The Court took note, for example, of testi-mony that patients and surgeons often re-quested specific anesthesiologists not asso-ciated with a hospital. Jefferson Parish,466 U.S. at 22, 104 S.Ct. 1551. Indirectevidence includes the behavior of firmswithout market power in the tying goodmarket, presumably on the notion that(competitive) supply follows demand. Ifcompetitive firms always bundle the tyingand tied goods, then they are a singleproduct. See id. at 22 n. 36, 104 S.Ct.1551; see also Eastman Kodak, 504 U.S.at 462, 112 S.Ct. 2072; Fortner I, 394 U.S.at 525, 89 S.Ct. 1252 (Fortas, J., dissent-ing), cited in Jefferson Parish, 466 U.S. at12, 22 n. 35, 104 S.Ct. 1551; United Statesv. Jerrold Elecs. Corp., 187 F.Supp. 545,559 (E.D.Pa.1960), aff’d per curiam, 365U.S. 567, 81 S.Ct. 755, 5 L.Ed.2d 806(1961); 10 PHILLIP E. AREEDA ET AL., ANTI-

TRUST LAW ¶ 1744, at 197–201 (1996). Herethe Court noted that only 27% of anesthe-siologists in markets other than the defen-dant’s had financial relationships with hos-pitals, and that, unlike radiologists andpathologists, anesthesiologists were notusually employed by hospitals, i.e., bundledwith hospital services. Jefferson Parish,466 U.S. at 22 n. 36, 104 S.Ct. 1551. With


both direct and indirect evidence concur-ring, the Court determined that hospitalsurgery and anesthesiological serviceswere distinct goods.

To understand the logic behind theCourt’s consumer demand test, considerfirst the postulated harms from tying.The core concern is that tying preventsgoods from competing directly for consum-er choice on their merits, i.e., being select-ed as a result of ‘‘buyers’ independentjudgment,’’ id. at 13, 104 S.Ct. 1551 (inter-nal quotes omitted). With a tie, a buyer’s‘‘freedom to select the best bargain in thesecond market [could be] impaired by hisneed to purchase the tying product, andperhaps by an inability to evaluate the truecost of either productTTTT’’ Id. at 15, 104S.Ct. 1551. Direct competition on themerits of the tied product is foreclosedwhen the tying product either is sold onlyin a bundle with the tied product or,though offered separately, is sold at a bun-dled price, so that the buyer pays the sameprice whether he takes the tied product ornot. In both cases, a consumer buying thetying product becomes entitled to the tiedproduct; he will therefore likely be unwill-ing to buy a competitor’s version of thetied product even if, making his ownprice/quality assessment, that is what hewould prefer.

But not all ties are bad. Bundling obvi-ously saves distribution and consumertransaction costs. 9 PHILLIP E. AREEDA,

ANTITRUST LAW ¶ 1703g2, at 51–52 (1991).This is likely to be true, to take someexamples from the computer industry, withthe integration of math co-processors andmemory into microprocessor chips and theinclusion of spell checkers in word pro-cessors. 11/10/98 pm Tr. at 18–19 (trialtestimony of Steven McGeady of Intel),reprinted in 9 J.A. at 5581–82 (math co-processor); Cal. Computer Prods., Inc. v.IBM Corp., 613 F.2d 727, 744 & n. 29 (9th

Cir.1979) (memory). Bundling can alsocapitalize on certain economies of scope.A possible example is the ‘‘shared’’ libraryfiles that perform OS and browser func-tions with the very same lines of code andthus may save drive space from the clutterof redundant routines and memory whenconsumers use both the OS and browsersimultaneously. 11/16/98 pm Tr. at 44 (tri-al testimony of Glenn Weadock), reprintedin 9 J.A. at 5892; Direct Testimony ofMicrosoft’s James Allchin ¶ ¶ 10, 97, 100,106–116, app. A (excluding ¶ ¶ f, g.vi), re-printed in 5 J.A. at 3292, 3322–30, 3412–17. Indeed, if there were no efficienciesfrom a tie (including economizing on con-sumer transaction costs such as the timeand effort involved in choice), we wouldexpect distinct consumer demand for eachindividual component of every good. In acompetitive market with zero transactioncosts, the computers on which this opinionwas written would only be sold piece-meal—keyboard, monitor, mouse, centralprocessing unit, disk drive, and memory allsold in separate transactions and likely bydifferent manufacturers.

Recognizing the potential benefits fromtying, see Jefferson Parish, 466 U.S. at 21n. 33, 104 S.Ct. 1551, the Court in Jeffer-son Parish forged a separate-products testthat, like those of market power and sub-stantial foreclosure, attempts to screen outfalse positives under per se analysis. Theconsumer demand test is a rough proxy forwhether a tying arrangement may, on bal-ance, be welfare-enhancing, and unsuitedto per se condemnation. In the abstract,of course, there is always direct separatedemand for products: assuming choice isavailable at zero cost, consumers will pre-fer it to no choice. Only when the efficien-cies from bundling are dominated by thebenefits to choice for enough consumers,however, will we actually observe consum-ers making independent purchases. Inother words, perceptible separate demand


is inversely proportional to net efficiencies.On the supply side, firms without marketpower will bundle two goods only when thecost savings from joint sale outweigh thevalue consumers place on separate choice.So bundling by all competitive firms im-plies strong net efficiencies. If a courtfinds either that there is no noticeableseparate demand for the tied product or,there being no convincing direct evidenceof separate demand, that the entire ‘‘com-petitive fringe’’ engages in the same be-havior as the defendant, 10 AREEDA ET AL.,

ANTITRUST LAW ¶ 1744c4, at 200, then thetying and tied products should be declaredone product and per se liability should berejected.

Before concluding our exegesis of Jeffer-son Parish’s separate-products test, weshould clarify two things. First, JeffersonParish does not endorse a direct inquiryinto the efficiencies of a bundle. Rather, itproposes easy-to-administer proxies fornet efficiency. In describing the separate-products test we discuss efficiencies onlyto explain the rationale behind the con-sumer demand inquiry. To allow the sepa-rate-products test to become a detailedinquiry into possible welfare consequenceswould turn a screening test into the veryprocess it is expected to render unneces-sary. 10 AREEDA ET AL., ANTITRUST LAW

¶ ¶ 1741b & c, at 180–85; see also Jeffer-son Parish, 466 U.S. at 34–35, 104 S.Ct.1551 (O’Connor, J., concurring).

Second, the separate-products test is nota one-sided inquiry into the cost savingsfrom a bundle. Although Jefferson Parishacknowledged that prior lower court caseslooked at cost-savings to decide separateproducts, see id. at 22 n. 35, 104 S.Ct. 1551,the Court conspicuously did not adopt thatapproach in its disposition of tying ar-rangement before it. Instead it choseproxies that balance costs savings againstreduction in consumer choice.

With this background, we now turn tothe separateproducts inquiry before us.The District Court found that many con-sumers, if given the option, would choosetheir browser separately from the OS.Findings of Fact ¶ 151 (noting that ‘‘corpo-rate consumers TTT prefer to standardizeon the same browser across different[OSs]’’ at the workplace). Turning to in-dustry custom, the court found that, al-though all major OS vendors bundledbrowsers with their OSs, these companieseither sold versions without a browser, orallowed OEMs or end-users either not toinstall the bundled browser or in any eventto ‘‘uninstall’’ it. Id. ¶ 153. The court didnot discuss the record evidence as towhether OS vendors other than Microsoftsold at a bundled price, with no discountfor a browserless OS, perhaps because therecord evidence on the issue was in con-flict. Compare, e.g., Direct Testimony ofRichard Schmalensee ¶ 241, reprinted in 7J.A. at 4315 (‘‘[A]ll major operating systemvendors do in fact include Web-browsingsoftware with the operating system at noextra charge.’’) (emphasis added), with,e.g., 1/6/99 pm Tr. at 42 (trial testimony ofFranklin Fisher of MIT) (suggesting allOSs but Microsoft offer discounts).

Microsoft does not dispute that manyconsumers demand alternative browsers.But on industry custom Microsoft contendsthat no other firm requires non-removalbecause no other firm has invested theresources to integrate web browsing asdeeply into its OS as Microsoft has. Ap-pellant’s Opening Br. at 25; cf. DirectTestimony of James Allchin ¶ ¶ 262–72, re-printed in 5 J.A. at 3385–89 (Apple, IBM);11/5/98 pm Tr. at 55–58 (trial testimony ofApple’s Avadis Tevanian, Jr.), reprinted in9 J.A. at 5507–10 (Apple). (We here usethe term ‘‘integrate’’ in the rather simplesense of converting individual goods intocomponents of a single physical object


(e.g., a computer as it leaves the OEM, ora disk or sets of disks), without any norma-tive implication that such integration isdesirable or achieves special advantages.Cf. United States v. Microsoft Corp., 147F.3d 935, 950 (D.C.Cir.1998) (‘‘MicrosoftII’’).) Microsoft contends not only that itsintegration of IE into Windows is innova-tive and beneficial but also that it requiresnon-removal of IE. In our discussion ofmonopoly maintenance we find that theseclaims fail the efficiency balancing applica-ble in that context. But the separate-products analysis is supposed to performits function as a proxy without embarkingon any direct analysis of efficiency. Ac-cordingly, Microsoft’s implicit argument—that in this case looking to a competitivefringe is inadequate to evaluate fully itspotentially innovative technological inte-gration, that such a comparison is betweenapples and oranges—poses a legitimate ob-jection to the operation of Jefferson Par-ish’s separate-products test for the per serule.

In fact there is merit to Microsoft’sbroader argument that Jefferson Parish’sconsumer demand test would ‘‘chill innova-tion to the detriment of consumers bypreventing firms from integrating intotheir products new functionality previouslyprovided by standalone products—andhence, by definition, subject to separateconsumer demand.’’ Appellant’s OpeningBr. at 69. The per se rule’s direct con-sumer demand and indirect industry cus-tom inquiries are, as a general matter,backward-looking and therefore systemat-ically poor proxies for overall efficiency inthe presence of new and innovative inte-gration. See 10 AREEDA ET AL., ANTITRUST

LAW ¶ 1746, at 224–29; Amicus Brief ofLawrence Lessig at 24–25, and sourcescited therein (brief submitted regardingConclusions of Law). The direct consumerdemand test focuses on historic consumerbehavior, likely before integration, and the

indirect industry custom test looks at firmsthat, unlike the defendant, may not haveintegrated the tying and tied goods. Bothtests compare incomparables—the defen-dant’s decision to bundle in the presence ofintegration, on the one hand, and consum-er and competitor calculations in its ab-sence, on the other. If integration hasefficiency benefits, these may be ignoredby the Jefferson Parish proxies. Becauseone cannot be sure beneficial integrationwill be protected by the other elements ofthe per se rule, simple application of thatrule’s separate-products test may makeconsumers worse off.

In light of the monopoly maintenancesection, obviously, we do not find that Mi-crosoft’s integration is welfare-enhancingor that it should be absolved of tying liabil-ity. Rather, we heed Microsoft’s warningthat the separate-products element of theper se rule may not give newly integratedproducts a fair shake.

B. Per Se Analysis Inappropriate forthis Case.

We now address directly the largerquestion as we see it: whether standardper se analysis should be applied ‘‘off theshelf’’ to evaluate the defendant’s tyingarrangement, one which involves softwarethat serves as a platform for third-partyapplications. There is no doubt that ‘‘[i]tis far too late in the history of our anti-trust jurisprudence to question the propo-sition that certain tying arrangementspose an unacceptable risk of stifling com-petition and therefore are unreasonable‘per se.’ ’’ Jefferson Parish, 466 U.S. at 9,104 S.Ct. 1551 (emphasis added). Butthere are strong reasons to doubt that theintegration of additional software function-ality into an OS falls among these arrange-ments. Applying per se analysis to suchan amalgamation creates undue risks of


error and of deterring welfare-enhancinginnovation.

The Supreme Court has warned that‘‘ ‘[i]t is only after considerable experiencewith certain business relationships thatcourts classify them as per se viola-tionsTTTT’ ’’ Broad. Music, 441 U.S. at 9,99 S.Ct. 1551 (quoting Topco Assocs., 405U.S. at 607–08, 92 S.Ct. 1126); accordCont’l T.V., Inc. v. GTE Sylvania Inc., 433U.S. 36, 47–59, 97 S.Ct. 2549, 53 L.Ed.2d568 (1977); White Motor Co. v. UnitedStates, 372 U.S. 253, 263, 83 S.Ct. 696, 9L.Ed.2d 738 (1963); Jerrold Elecs., 187F.Supp. at 555–58, 560–61; see also FrankH. Easterbrook, Allocating Antitrust De-cisionmaking Tasks, 76 GEO. L.J. 305, 308(1987). Yet the sort of tying arrangementattacked here is unlike any the SupremeCourt has considered. The early SupremeCourt cases on tying dealt with arrange-ments whereby the sale or lease of a pat-ented product was conditioned on the pur-chase of certain unpatented products fromthe patentee. See Motion Picture Pat-ents, 243 U.S. 502, 37 S.Ct. 416, 61 L.Ed.871 (1917); United Shoe Mach., 258 U.S.451, 42 S.Ct. 363, 66 L.Ed. 708 (1922);IBM Corp. v. United States, 298 U.S. 131,56 S.Ct. 701, 80 L.Ed. 1085 (1936); Int’lSalt, 332 U.S. 392, 68 S.Ct. 12, 92 L.Ed. 20(1947). Later Supreme Court tying casesdid not involve market power derived frompatents, but continued to involve contrac-tual ties. See Times–Picayune, 345 U.S.594, 73 S.Ct. 872, 97 L.Ed. 1277 (1953)(defendant newspaper conditioned the pur-chase of ads in its evening edition on thepurchase of ads in its morning edition); N.Pac. Ry., 356 U.S. 1, 78 S.Ct. 514, 2L.Ed.2d 545 (1958) (defendant railroadleased land only on the condition thatproducts manufactured on the land beshipped on its railways); United States v.Loew’s Inc., 371 U.S. 38, 83 S.Ct. 97, 9L.Ed.2d 11 (1962) (defendant distributor ofcopyrighted feature films conditioned the

sale of desired films on the purchase ofundesired films); U.S. Steel Corp. v. Fort-ner Enters., Inc., 429 U.S. 610, 97 S.Ct.861, 51 L.Ed.2d 80 (1977) (‘‘Fortner II’’)(defendant steel company conditioned ac-cess to low interest loans on the purchaseof the defendant’s prefabricated homes);Jefferson Parish, 466 U.S. 2, 104 S.Ct.1551, 80 L.Ed.2d 2 (1984) (defendant hos-pital conditioned use of its operating roomson the purchase of anesthesiological ser-vices from a medical group associated withthe hospital); Eastman Kodak, 504 U.S.451, 112 S.Ct. 2072, 119 L.Ed.2d 265 (1992)(defendant photocopying machine manu-facturer conditioned the sale of replace-ment parts for its machines on the use ofthe defendant’s repair services).

In none of these cases was the tied goodphysically and technologically integratedwith the tying good. Nor did the defen-dants ever argue that their tie improvedthe value of the tying product to users andto makers of complementary goods. Inthose cases where the defendant claimedthat use of the tied good made the tyinggood more valuable to users, the Courtruled that the same result could beachieved via quality standards for substi-tutes of the tied good. See, e.g., Int’l Salt,332 U.S. at 397–98, 68 S.Ct. 12; IBM, 298U.S. at 138–40, 56 S.Ct. 701. Here Micro-soft argues that IE and Windows are anintegrated physical product and that thebundling of IE APIs with Windows makesthe latter a better applications platform forthird-party software. It is unclear howthe benefits from IE APIs could beachieved by quality standards for differentbrowser manufacturers. We do not passjudgment on Microsoft’s claims regardingthe benefits from integration of its APIs.We merely note that these and other nov-el, purported efficiencies suggest that judi-cial ‘‘experience’’ provides little basis forbelieving that, ‘‘because of their pernicious


effect on competition and lack of any re-deeming virtue,’’ a software firm’s deci-sions to sell multiple functionalities as apackage should be ‘‘conclusively presumedto be unreasonable and therefore illegalwithout elaborate inquiry as to the preciseharm they have caused or the businessexcuse for their use.’’ N. Pac. Ry., 356U.S. at 5, 78 S.Ct. 514 (emphasis added).

Nor have we found much insight intosoftware integration among the decisionsof lower federal courts. Most tying casesin the computer industry involve bundlingwith hardware. See, e.g., Digital Equip.Corp. v. Uniq Digital Techs., Inc., 73 F.3d756, 761 (7th Cir.1996) (Easterbrook, J.)(rejecting with little discussion the notionthat bundling of OS with a computer is atie of two separate products); Datagate,Inc. v. Hewlett–Packard Co., 941 F.2d 864,870 (9th Cir.1991) (holding that plaintiff’sallegation that defendant conditioned itssoftware on purchase of its hardware wassufficient to survive summary judgment);Digidyne Corp. v. Data Gen. Corp., 734F.2d 1336, 1341–47 (9th Cir.1984) (holdingthat defendant’s conditioning the sale of itsOS on the purchase of its CPU constitutesa per se tying violation); Cal. ComputerProds., 613 F.2d at 743–44 (holding thatdefendant’s integration into its CPU of adisk controller designed for its own diskdrives was a useful innovation and not animpermissible attempt to monopolize);ILC Peripherals Leasing Corp. v. IBMCorp., 448 F.Supp. 228, 233 (N.D.Cal.1978)(finding that defendant’s integration ofmagnetic disks and a head/disk assemblywas not an unlawful tie), aff’d per curiamsub. nom. Memorex Corp. v. IBM Corp.,636 F.2d 1188 (9th Cir.1980); see alsoTransamerica Computer Co. v. IBMCorp., 698 F.2d 1377, 1382–83 (9th Cir.1983) (finding lawful defendant’s designchanges that rendered plaintiff peripheralmaker’s tape drives incompatible with thedefendant’s CPU). The hardware case

that most resembles the present one isTelex Corp. v. IBM Corp., 367 F.Supp. 258(N.D.Okla.1973), rev’d on other grounds,510 F.2d 894 (10th Cir.1975). Just as Mi-crosoft integrated web browsing into itsOS, IBM in the 1970s integrated memoryinto its CPUs, a hardware platform. Aperipheral manufacturer alleged a tyingviolation, but the District Court dismissedthe claim because it thought it inappropri-ate to enmesh the courts in product designdecisions. Id. at 347. The court’s discus-sion of the tying claim was brief and didnot dwell on the effects of the integrationon competition or efficiencies. Nor did thecourt consider whether per se analysis ofthe alleged tie was wise.

We have found four antitrust cases in-volving arrangements in which a softwareprogram is tied to the purchase of a soft-ware platform—two district court casesand two appellate court cases, includingone from this court. The first case, Inno-vation Data Processing, Inc. v. IBMCorp., 585 F.Supp. 1470 (D.N.J.1984), in-volved an allegation that IBM bundledwith its OS a utility used to transfer datafrom a tape drive to a computer’s diskdrive. Although the court mentioned theefficiencies achieved by bundling, it ulti-mately dismissed the per se tying claimbecause IBM sold a discounted version ofthe OS without the utility. Id. at 1475–76.The second case, A.I. Root Co. v. Comput-er/Dynamics, Inc., 806 F.2d 673 (6th Cir.1986), was brought by a business customerwho claimed that an OS manufacturer ille-gally conditioned the sale of its OS on thepurchase of other software applications.The court quickly disposed of the case onthe ground that defendant Computer/Dy-namics had no market power. Id. at 675–77. There was no mention of the efficien-cies from the tie. The third case, Caldera,Inc. v. Microsoft Corp., 72 F.Supp.2d 1295


(D.Utah 1999), involved a complaint thatthe technological integration of MS–DOSand Windows 3.1 into Windows 95 consti-tuted a per se tying violation. The courtformulated the ‘‘single product’’ issue interms of whether the tie constituted atechnological improvement, ultimately con-cluding that Microsoft was not entitled tosummary judgment on that issue. Id. at1322–28.

The software case that bears the great-est resemblance to that at bar is, not sur-prisingly, Microsoft II, 147 F.3d 935,where we examined the bundling of IEwith Windows 95. But the issue there waswhether the bundle constituted an ‘‘inte-grated product’’ as the term was used in a1994 consent decree between the Depart-ment of Justice and Microsoft. Id. at 939.We did not consider whether Microsoft’sbundling should be condemned as per seillegal. We certainly did not make anyfinding that bundling IE with Windowshad ‘‘no purpose except stifling of competi-tion,’’ White Motor, 372 U.S. at 263, 83S.Ct. 696, an important consideration indefining the scope of any of antitrust law’sper se rules, see Cont’l T.V., 433 U.S. at57–59, 97 S.Ct. 2549. While we believedour interpretation of the term ‘‘integratedproduct’’ was consistent with the test forseparate products under tying law, wemade clear that the ‘‘antitrust question isof course distinct.’’ Microsoft II, 147 F.3dat 950 n. 14. We even cautioned that ourconclusion that IE and Windows 95 wereintegrated was ‘‘subject to reexaminationon a more complete record.’’ Id. at 952.To the extent that the decision completelydisclaimed judicial capacity to evaluate‘‘high-tech product design,’’ id., it cannotbe said to conform to prevailing antitrustdoctrine (as opposed to resolution of thedecree-interpretation issue then before us).In any case, mere review of assertedbreaches of a consent decree hardly consti-tutes enough ‘‘experience’’ to warrant ap-

plication of per se analysis. See Broad.Music, 441 U.S. at 10–16, 99 S.Ct. 1551(refusing to apply per se analysis to defen-dant’s blanket licenses even though thoselicenses had been thoroughly investigatedby the Department of Justice and were thesubject of a consent decree that had beenreviewed by numerous courts).

While the paucity of cases examiningsoftware bundling suggests a high riskthat per se analysis may produce inaccu-rate results, the nature of the platformsoftware market affirmatively suggeststhat per se rules might stunt valuable in-novation. We have in mind two reasons.

First, as we explained in the previoussection, the separate-products test is apoor proxy for net efficiency from newlyintegrated products. Under per se analy-sis the first firm to merge previously dis-tinct functionalities (e.g., the inclusion ofstarter motors in automobiles) or to elimi-nate entirely the need for a second func-tion (e.g., the invention of the stain-resis-tant carpet) risks being condemned ashaving tied two separate products becauseat the moment of integration there will ap-pear to be a robust ‘‘distinct’’ market forthe tied product. See 10 AREEDA ET AL.,

ANTITRUST LAW ¶ 1746, at 224. Rule ofreason analysis, however, affords the firstmover an opportunity to demonstrate thatan efficiency gain from its ‘‘tie’’ adequatelyoffsets any distortion of consumer choice.See Grappone, Inc. v. Subaru of New Eng-land, Inc., 858 F.2d 792, 799 (1st Cir.1988)(Breyer, J.); see also Town Sound & Cus-tom Tops, Inc. v. Chrysler Motor Corp.,959 F.2d 468, 482 (3d Cir.1992); KaiserAluminum & Chem. Sales, Inc. v. Avon-dale Shipyards, Inc., 677 F.2d 1045, 1048–49 n. 5 (5th Cir.1982).

The failure of the separate-products testto screen out certain cases of productiveintegration is particularly troubling in plat-


form software markets such as that inwhich the defendant competes. Not onlyis integration common in such markets,but it is common among firms withoutmarket power. We have already reviewedevidence that nearly all competitive OSvendors also bundle browsers. Moreover,plaintiffs do not dispute that OS vendorscan and do incorporate basic internetplumbing and other useful functionalityinto their OSs. See Direct Testimony ofRichard Schmalensee ¶ 508, reprinted in 7J.A. at 4462–64 (disk defragmentation,memory management, peer-to-peer net-working or file sharing); 11/19/98 am Tr.at 82–83 (trial testimony of FrederickWarren–Boulton), reprinted in 10 J.A. at6427–28 (TCP/IP stacks). Firms withoutmarket power have no incentive to pack-age different pieces of software togetherunless there are efficiency gains from do-ing so. The ubiquity of bundling in com-petitive platform software markets shouldgive courts reason to pause before con-demning such behavior in less competitivemarkets.

Second, because of the pervasively inno-vative character of platform software mar-kets, tying in such markets may produceefficiencies that courts have not previouslyencountered and thus the Supreme Courthad not factored into the per se rule asoriginally conceived. For example, thebundling of a browser with OSs enables anindependent software developer to counton the presence of the browser’s APIs, ifany, on consumers’ machines and thus toomit them from its own package. SeeDirect Testimony of Richard Schmalensee¶ ¶ 230–31, 234, reprinted in 7 J.A. at4309–11, 4312; Direct Testimony of Mi-chael Devlin ¶ ¶ 12–21, reprinted in 5 J.A.at 3525–29; see also Findings of Fact ¶ 2.It is true that software developers canbundle the browser APIs they need withtheir own products, see id. ¶ 193, but thatmay force consumers to pay twice for the

same API if it is bundled with two differ-ent software programs. It is also truethat OEMs can include APIs with the com-puters they sell, id., but diffusion of uni-form APIs by that route may be inferior.First, many OEMs serve special subsets ofWindows consumers, such as home or cor-porate or academic users. If just one ofthese OEMs decides not to bundle an APIbecause it does not benefit enough of itsclients, ISVs that use that API might haveto bundle it with every copy of their pro-gram. Second, there may be a substantiallag before all OEMs bundle the same setof APIs—a lag inevitably aggravated bythe first phenomenon. In a field whereprograms change very rapidly, delays inthe spread of a necessary element (here,the APIs) may be very costly. Of course,these arguments may not justify Micro-soft’s decision to bundle APIs in this case,particularly because Microsoft did notmerely bundle with Windows the APIsfrom IE, but an entire browser application(sometimes even without APIs, see id.). Ajustification for bundling a component ofsoftware may not be one for bundling theentire software package, especially giventhe malleability of software code. See id.¶ ¶ 162–63; 12/9/98 am Tr. at 17 (trial tes-timony of David Farber); 1/6/99 am Tr. at6–7 (trial testimony of Franklin Fisher),reprinted in 11 J.A. at 7192–93; DirectTestimony of Joachim Kempin ¶ 286, re-printed in 6 J.A. at 3749. Furthermore,the interest in efficient API diffusion obvi-ously supplies a far stronger justificationfor simple price-bundling than for Micro-soft’s contractual or technological bars tosubsequent removal of functionality. Butour qualms about redefining the bound-aries of a defendant’s product and thepossibility of consumer gains from simpli-fying the work of applications developersmakes us question any hard and fast ap-proach to tying in OS software markets.


There may also be a number of efficien-cies that, although very real, have beenignored in the calculations underlying theadoption of a per se rule for tying. Wefear that these efficiencies are common intechnologically dynamic markets whereproduct development is especially unlikelyto follow an easily foreseen linear pattern.Take the following example from ILC Pe-ripherals, 448 F.Supp. 228, a case concern-ing the evolution of disk drives for comput-ers. When IBM first introduced suchdrives in 1956, it sold an integrated prod-uct that contained magnetic disks and diskheads that read and wrote data onto disks.Id. at 231. Consumers of the drives de-manded two functions—to store data andto access it all at once. In the first fewyears consumers’ demand for storage in-creased rapidly, outpacing the evolution ofmagnetic disk technology. To satisfy thatdemand IBM made it possible for consum-ers to remove the magnetic disks fromdrives, even though that meant consumerswould not have access to data on disksremoved from the drive. This componen-tization enabled makers of computer pe-ripherals to sell consumers removabledisks. Id. at 231–32. Over time, however,the technology of magnetic disks caught upwith demand for capacity, so that consum-ers needed few removable disks to storeall their data. At this point IBM reinte-grated disks into their drives, enablingconsumers to once again have immediateaccess to all their data without a sacrificein capacity. Id. A manufacturer of re-movable disks sued. But the DistrictCourt found the tie justified because itsatisfied consumer demand for immediateaccess to all data, and ruled that disks anddisk heads were one product. Id. at 233.A court hewing more closely to the trun-cated analysis contemplated by NorthernPacific Railway would perhaps have over-looked these consumer benefits.

These arguments all point to one conclu-sion: we cannot comfortably say that bun-dling in platform software markets has solittle ‘‘redeeming virtue,’’ N. Pac. Ry., 356U.S. at 5, 78 S.Ct. 514, and that therewould be so ‘‘very little loss to society’’from its ban, that ‘‘an inquiry into its costsin the individual case [can be] considered[ ] unnecessary.’’ Jefferson Parish, 466U.S. at 33–34, 104 S.Ct. 1551 (O’Connor, J.,concurring). We do not have enough em-pirical evidence regarding the effect of Mi-crosoft’s practice on the amount of con-sumer surplus created or consumer choiceforeclosed by the integration of addedfunctionality into platform software to ex-ercise sensible judgment regarding thatentire class of behavior. (For some issueswe have no data.) ‘‘We need to knowmore than we do about the actual impactof these arrangements on competition todecide whether they TTT should be classi-fied as per se violations of the ShermanAct.’’ White Motor, 372 U.S. at 263, 83S.Ct. 696. Until then, we will heed thewisdom that ‘‘easy labels do not alwayssupply ready answers,’’ Broad. Music, 441U.S. at 8, 99 S.Ct. 1551, and vacate theDistrict Court’s finding of per se tyingliability under Sherman Act § 1. We re-mand the case for evaluation of Microsoft’stying arrangements under the rule of rea-son. See Pullman–Standard v. Swint, 456U.S. 273, 292, 102 S.Ct. 1781, 72 L.Ed.2d66 (1982) (‘‘[W]here findings are infirmbecause of an erroneous view of the law, aremand is the proper course unless therecord permits only one resolution of thefactual issue.’’). That rule more freelypermits consideration of the benefits ofbundling in software markets, particularlythose for OSs, and a balancing of thesebenefits against the costs to consumerswhose ability to make direct price/qualitytradeoffs in the tied market may havebeen impaired. See Jefferson Parish, 466U.S. at 25 nn.41–42, 104 S.Ct. 1551 (noting


that per se rule does not broadly permitconsideration of procompetitive justifica-tions); id. at 34–35, 104 S.Ct. 1551 (O’Con-nor, J., concurring); N. Pac. Ry., 356 U.S.at 5, 78 S.Ct. 514.

Our judgment regarding the compara-tive merits of the per se rule and the ruleof reason is confined to the tying arrange-ment before us, where the tying product issoftware whose major purpose is to serveas a platform for third-party applicationsand the tied product is complementarysoftware functionality. While our reason-ing may at times appear to have broaderforce, we do not have the confidence tospeak to facts outside the record, whichcontains scant discussion of software inte-gration generally. Microsoft’s primaryjustification for bundling IE APIs is thattheir inclusion with Windows increases thevalue of third-party software (and Win-dows) to consumers. See Appellant’sOpening Br. at 41–43. Because this claimapplies with distinct force when the tyingproduct is platform software, we have nopresent basis for finding the per se ruleinapplicable to software markets generally.Nor should we be interpreted as setting aprecedent for switching to the rule of rea-son every time a court identifies an effi-ciency justification for a tying arrange-ment. Our reading of the record suggestsmerely that integration of new functionali-ty into platform software is a commonpractice and that wooden application of perse rules in this litigation may cast a cloudover platform innovation in the market forPCs, network computers and informationappliances.

C. On Remand

Should plaintiffs choose to pursue a ty-ing claim under the rule of reason, we notethe following for the benefit of the trialcourt:

[55] First, on remand, plaintiffs mustshow that Microsoft’s conduct unreason-ably restrained competition. Meeting thatburden ‘‘involves an inquiry into the actualeffect’’ of Microsoft’s conduct on competi-tion in the tied good market, JeffersonParish, 466 U.S. at 29, 104 S.Ct. 1551, theputative market for browsers. To the ex-tent that certain aspects of tying injurymay depend on a careful definition of thetied good market and a showing of barri-ers to entry other than the tying arrange-ment itself, plaintiffs would have to estab-lish these points. See Jefferson Parish,466 U.S. at 29, 104 S.Ct. 1551 (‘‘This com-petition [among anesthesiologists] takesplace in a market that has not been de-fined.’’); id. at 29 n. 48, 104 S.Ct. 1551(‘‘[N]either the District Court nor theCourt of Appeals made any findings con-cerning the contract’s effect on entry bar-riers.’’). But plaintiffs were required—andhad every incentive—to provide both adefinition of the browser market and barri-ers to entry to that market as part of their§ 2 attempted monopolization claim; yetthey failed to do so. See supra SectionIII. Accordingly, on remand of the § 1tying claim, plaintiffs will be precludedfrom arguing any theory of harm thatdepends on a precise definition of browsersor barriers to entry (for example, networkeffects from Internet protocols and exten-sions embedded in a browser) other thanwhat may be implicit in Microsoft’s tyingarrangement.

Of the harms left, plaintiffs must showthat Microsoft’s conduct was, on balance,anticompetitive. Microsoft may of courseoffer procompetitive justifications, and it isplaintiffs’ burden to show that the anticom-petitive effect of the conduct outweighs itsbenefit.

[56] Second, the fact that we have al-ready considered some of the behaviorplaintiffs allege to constitute tying viola-


tions in the monopoly maintenance sectiondoes not resolve the § 1 inquiry. The twopractices that plaintiffs have most ardent-ly claimed as tying violations are, indeed,a basis for liability under plaintiffs’ § 2monopoly maintenance claim. These areMicrosoft’s refusal to allow OEMs to unin-stall IE or remove it from the Windowsdesktop, Findings of Fact ¶ ¶ 158, 203,213, and its removal of the IE entry fromthe Add/Remove Programs utility in Win-dows 98, id. ¶ 170. See supra SectionII.B. In order for the District Court toconclude these practices also constitute§ 1 tying violations, plaintiffs must dem-onstrate that their benefits—if any, seesupra Sections II.B.1.b and II.B.2.b;Findings of Fact ¶ ¶ 176, 186, 193—areoutweighed by the harms in the tied prod-uct market. See Jefferson Parish, 466U.S. at 29, 104 S.Ct. 1551. If the DistrictCourt is convinced of net harm, it mustthen consider whether any additional rem-edy is necessary.

In Section II.B we also considered an-other alleged tying violation—the Windows98 override of a consumer’s choice of de-fault web browser. We concluded thatthis behavior does not provide a distinctbasis for § 2 liability because plaintiffsfailed to rebut Microsoft’s proffered justifi-cation by demonstrating that harms in theoperating system market outweigh Micro-soft’s claimed benefits. See supra SectionII.B. On remand, however, although Mi-crosoft may offer the same procompetitivejustification for the override, plaintiffsmust have a new opportunity to rebut thisclaim, by demonstrating that the anticom-petitive effect in the browser market isgreater than these benefits.

[57] Finally, the District Court mustalso consider an alleged tying violationthat we did not consider under § 2 monop-oly maintenance: price bundling. First,the court must determine if Microsoft in-

deed price bundled—that is, was Micro-soft’s charge for Windows and IE higherthan its charge would have been for Win-dows alone? This will require plaintiffs toresolve the tension between Findings ofFact ¶ ¶ 136–37, which Microsoft interpretsas saying that no part of the bundled priceof Windows can be attributed to IE, andConclusions of Law, at 50, which says theopposite. Compare Direct Testimony ofPaul Maritz ¶ ¶ 37, 296, reprinted in 6 J.A.at 3656, 3753–54 (Microsoft did not‘‘charge separately’’ for IE, but like allother major OS vendors included browsingsoftware at ‘‘no extra charge’’), with GX202 at MS7 004343, esp. 004347, reprintedin 22 J.A. at 14459, esp. 14463 (memo fromChristian Wildfeuer describing focus grouptest used to price Windows 98 with IE 4),and GX 1371 at MS7 003729–30, 003746,003748, esp. 003750, reprinted in 15 J.A. at10306–07, 10323, 10325, esp. 10327 (Win-dows 98 pricing and marketing memo),and Findings of Fact ¶ 63 (identifying GX202 as the basis for Windows 98 pricing).

If there is a positive price increment inWindows associated with IE (we knowthere is no claim of price predation),plaintiffs must demonstrate that the anti-competitive effects of Microsoft’s pricebundling outweigh any procompetitive jus-tifications the company provides for it. Instriking this balance, the District Courtshould consider, among other things, indi-rect evidence of efficiency provided by‘‘the competitive fringe.’’ See supra Sec-tion IV.A. Although this inquiry may ov-erlap with the separate-products screenunder the per se rule, that is not its rolehere. Because courts applying the rule ofreason are free to look at both direct andindirect evidence of efficiencies from a tie,there is no need for a screening device assuch; thus the separate-products inquiryserves merely to classify arrangements assubject to tying law, as opposed to, say,


liability for exclusive dealing. See Times–Picayune, 345 U.S. at 614, 73 S.Ct. 872(finding a single product and then turningto a general rule of reason analysis under§ 1, though not using the term ‘‘tying’’);Foster v. Md. State Sav. & Loan Ass’n,590 F.2d 928, 931, 933 (D.C.Cir.1978), cit-ed in Jefferson Parish, 466 U.S. at 40, 104S.Ct. 1551 (O’Connor, J., concurring)(same); see also Chawla v. Shell Oil Co.,75 F.Supp.2d 626, 635, 643–44 (S.D.Tex.1999) (considering a rule of reason tyingclaim after finding a single product underthe per se rule); Montgomery CountyAss’n of Realtors v. Realty Photo MasterCorp., 783 F.Supp. 952, 961 & n. 26(D.Md.1992), aff’d mem. 993 F.2d 1538(4th Cir.1993) (same).

If OS vendors without market poweralso sell their software bundled with abrowser, the natural inference is that saleof the items as a bundle serves consumerdemand and that unbundled sale wouldnot, for otherwise a competitor could prof-itably offer the two products separatelyand capture sales of the tying good fromvendors that bundle. See 10 AREEDA ET AL.,

ANTITRUST LAW ¶ 1744b, at 197–98. It doesappear that most if not all firms have solda browser with their OSs at a bundledprice, beginning with IBM and its OS/2Warp OS in September 1994, Findings ofFact ¶ 140; see also Direct Testimony ofRichard Schmalensee ¶ 212, reprinted in 7J.A. at 4300–01, and running to currentversions of Apple’s Mac OS, Caldera andRed Hat’s Linux OS, Sun’s Solaris OS,Be’s BeOS, Santa Cruz Operation’s Unix-Ware, Novell’s NetWare OS, and others,see Findings of Fact ¶ 153; Direct Testi-mony of Richard Schmalensee ¶ ¶ 215–23,230, esp. table 5, reprinted in 7 J.A. at4302–05, 4310; Direct Testimony of JamesAllchin ¶ ¶ 261–77, reprinted in 5 J.A. at3384–92.

Of course price bundling by competitiveOS makers would tend to exonerate Micro-soft only if the sellers in question soldtheir browser/OS combinations exclusivelyat a bundled price. If a competitive selleroffers a discount for a browserless version,then—at least as to its OS and browser—the gains from bundling are outweighed bythose from separate choice. The evidenceon discounts appears to be in conflict.Compare Direct Testimony of RichardSchmalensee ¶ 241, reprinted in 7 J.A. at4315, with 1/6/99 pm Tr. at 42 (trial testi-mony of Franklin Fisher). If Schmalenseeis correct that nearly all OS makers do notoffer a discount, then the harm from ty-ing—obstruction of direct consumerchoice—would be theoretically created byvirtually all sellers: a customer who wouldprefer an alternate browser is forced topay the full price of that browser eventhough its value to him is only the incre-ment in value over the bundled browser.(The result is similar to that from non-removal, which forces consumers who wantthe alternate browser to surrender diskspace taken up by the unused, bundledbrowser.) If the failure to offer a pricediscount were universal, any impedimentto direct consumer choice created by Mi-crosoft’s price-bundled sale of IE withWindows would be matched throughoutthe market; yet these OS suppliers on thecompetitive fringe would have evidentlyfound this price bundling on balance effi-cient. If Schmalensee’s assertions are ill-founded, of course, no such inference couldbe drawn.

V. TRIAL PROCEEDINGS AND REMEDY

Microsoft additionally challenges theDistrict Court’s procedural rulings on twofronts. First, with respect to the trialphase, Microsoft proposes that the courtmismanaged its docket by adopting an ex-pedited trial schedule and receiving evi-dence through summary witnesses. Sec-


ond, with respect to the remedies decree,Microsoft argues that the court improperlyordered that it be divided into two sepa-rate companies. Only the latter claim willlong detain us. The District Court’s trial-phase procedures were comfortably withinthe bounds of its broad discretion to con-duct trials as it sees fit. We conclude,however, that the District Court’s reme-dies decree must be vacated for three in-dependent reasons: (1) the court failed tohold a remedies-specific evidentiary hear-ing when there were disputed facts; (2)the court failed to provide adequate rea-sons for its decreed remedies; and (3) thisCourt has revised the scope of Microsoft’sliability and it is impossible to determineto what extent that should affect the reme-dies provisions.

A. Factual Background

On April 3, 2000, the District Court con-cluded the liability phase of the proceed-ings by the filing of its Conclusions of Lawholding that Microsoft had violated §§ 1and 2 of the Sherman Act. The court andthe parties then began discussions of theprocedures to be followed in the impositionof remedies. Initially, the District Courtsignaled that it would enter relief onlyafter conducting a new round of proceed-ings. In its Conclusions of Law, the courtstated that it would issue a remedies order‘‘following proceedings to be established byfurther Order of the Court.’’ Conclusionsof Law, at 57. And, when during a post-trial conference, Microsoft’s counsel askedwhether the court ‘‘contemplate[d] furtherproceedings,’’ the judge replied, ‘‘Yes.Yes. I assume that there would be furtherproceedings.’’ 4/4/00 Tr. at 8–9, 11, re-printed in 4 J.A. at 2445–46, 2448. TheDistrict Court further speculated thatthose proceedings might ‘‘replicate theprocedure at trial with testimony in writ-ten form subject to crossexamination.’’ Id.at 11, reprinted in 4 J.A. at 2448.

On April 28, 2000, plaintiffs submittedtheir proposed final judgment, accompa-nied by six new supporting affidavits andseveral exhibits. In addition to a series oftemporary conduct restrictions, plaintiffsproposed that Microsoft be split into twoindependent corporations, with one con-tinuing Microsoft’s operating systems busi-ness and the other undertaking the bal-ance of Microsoft’s operations. Plaintiffs’Proposed Final Judgment at 2–3, reprint-ed in 4 J.A. at 2473–74. Microsoft filed a‘‘summary response’’ on May 10, contend-ing both that the proposed decree was toosevere and that it would be impossible toresolve certain remedies-specific factualdisputes ‘‘on a highly expedited basis.’’Defendant’s Summary Response at 6–7,reprinted in 4 J.A. at 2587–88. AnotherMay 10 submission argued that if the Dis-trict Court considered imposing plaintiffs’proposed remedy, ‘‘then substantial discov-ery, adequate time for preparation and afull trial on relief will be required.’’ De-fendant’s Position as to Future Proceed-ings at 2, reprinted in 4 J.A. at 2646.

After the District Court revealed duringa May 24 hearing that it was prepared toenter a decree without conducting ‘‘anyfurther process,’’ 5/24/00 pm Tr. at 33,reprinted in 14 J.A. at 9866, Microsoftrenewed its argument that the underlyingfactual disputes between the parties neces-sitated a remedies-specific evidentiaryhearing. In two separate offers of proof,Microsoft offered to produce a number ofpieces of evidence, including the following:

1 Testimony from Dr. Robert Cran-dall, a Senior Fellow at the Brook-ings Institution, that divestiture anddissolution orders historically have‘‘failed to improve economic welfareby reducing prices or increasing out-put.’’ Defendant’s Offer of Proof at2, reprinted in 4 J.A. at 2743.


1 Testimony from Professor KennethElzinga, Professor of Economics atthe University of Virginia, that plain-tiffs’ proposed remedies would notinduce entry into the operating sys-tems market. Id. at 4, reprinted in4 J.A. at 2745.

1 Testimony from Dean RichardSchmalensee, Dean of MIT’s SloanSchool of Management, that dividingMicrosoft likely would ‘‘harm con-sumers through higher prices, loweroutput, reduced efficiency, and lessinnovation’’ and would ‘‘produce im-mediate, substantial increases in theprices of both Windows and Office.’’Id. at 8, reprinted in 4 J.A. at 2749.Indeed, it would cause the price ofWindows to triple. Id.

1 Testimony from Goldman, Sachs &Co. and from Morgan Stanley DeanWitter that dissolution would ad-versely affect shareholder value. Id.at 17, 19, reprinted in 4 J.A. at 2758,2760.

1 Testimony from Microsoft ChairmanBill Gates that dividing Microsoft‘‘along the arbitrary lines proposedby the Government’’ would devastatethe company’s proposed Next Gener-ation Windows Services platform,which would allow software develop-ers to write web-based applicationsthat users could access from a widerange of devices. Id. at 21–22, re-printed in 4 J.A. at 2762–63.

1 Testimony from Steve Ballmer, Mi-crosoft’s President and CEO, thatMicrosoft is organized as a unifiedcompany and that ‘‘there are no nat-ural lines along which Microsoftcould be broken up without causingserious problems.’’ Id. at 23, re-printed in 4 J.A. at 2764.

1 Testimony from Michael Capellas,CEO of Compaq, that splitting Mi-

crosoft in two ‘‘will make it moredifficult for OEMs to provide cus-tomers with the tightly integratedproduct offerings they demand’’ inpart because ‘‘complementary prod-ucts created by unrelated companiesdo not work as well together as prod-ucts created by a single company.’’Defendant’s Supplemental Offer ofProof at 2, reprinted in 4 J.A. at2823.

Over Microsoft’s objections, the DistrictCourt proceeded to consider the merits ofthe remedy and on June 7, 2000 enteredits final judgment. The court explainedthat it would not conduct ‘‘extended pro-ceedings on the form a remedy shouldtake,’’ because it doubted that an eviden-tiary hearing would ‘‘give any significantlygreater assurance that it will be able toidentify what might be generally regardedas an optimum remedy.’’ Final Judg-ment, at 62. The bulk of Microsoft’s prof-fered facts were simply conjectures aboutfuture events, and ‘‘[i]n its experience theCourt has found testimonial predictions offuture events generally less reliable eventhan testimony as to historical fact, andcrossexamination to be of little use in en-hancing or detracting from their accura-cy.’’ Id. Nor was the court swayed byMicrosoft’s ‘‘profession of surprise’’ at thepossibility of structural relief. Id. at 61.‘‘From the inception of this case Microsoftknew, from well-established SupremeCourt precedents dating from the begin-ning of the last century, that a mandateddivestiture was a possibility, if not a proba-bility, in the event of an adverse result attrial.’’ Id.

The substance of the District Court’sremedies order is nearly identical to plain-tiffs’ proposal. The decree’s centerpiece isthe requirement that Microsoft submit aproposed plan of divestiture, with the com-pany to be split into an ‘‘Operating Sys-


tems Business,’’ or ‘‘OpsCo,’’ and an ‘‘Ap-plications Business,’’ or ‘‘AppsCo.’’ FinalJudgment, Decree §§ 1.a, l.c.i, at 64. Op-sCo would receive all of Microsoft’s operat-ing systems, such as Windows 98 and Win-dows 2000, while AppsCo would receivethe remainder of Microsoft’s businesses,including IE and Office. The DistrictCourt identified four reasons for its ‘‘reluc-tant[ ]’’ conclusion that ‘‘a structural reme-dy has become imperative.’’ Id. at 62.First, Microsoft ‘‘does not yet concede thatany of its business practices violated theSherman Act.’’ Id. Second, the companyconsequently ‘‘continues to do business asit has in the past.’’ Id. Third, Microsoft‘‘has proved untrustworthy in the past.’’Id. And fourth, the Government, whoseofficials ‘‘are by reason of office obligedand expected to consider—and to act in—the public interest,’’ won the case, ‘‘and forthat reason alone have some entitlement toa remedy of their choice.’’ Id. at 62–63.

The decree also contains a number ofinterim restrictions on Microsoft’s conduct.For instance, Decree § 3.b requires Micro-soft to disclose to third-party developersthe APIs and other technical informationnecessary to ensure that software effec-tively interoperates with Windows. Id. at67. ‘‘To facilitate compliance,’’ § 3.b fur-ther requires that Microsoft establish ‘‘asecure facility’’ at which third-party repre-sentatives may ‘‘study, interrogate and in-teract with relevant and necessary por-tions of [Microsoft platform software]source code.’’ Id. Section 3.e, entitled‘‘Ban on Exclusive Dealing,’’ forbids Micro-soft from entering contracts which obligethird parties to restrict their ‘‘develop-ment, production, distribution, promotionor use of, or payment for’’ non-Microsoftplatformlevel software. Id. at 68. UnderDecree § 3.f—‘‘Ban on Contractual Ty-ing’’—the company may not condition itsgrant of a Windows license on a party’sagreement ‘‘to license, promote, or distrib-

ute any other Microsoft software product.’’Id. And § 3.g imposes a ‘‘Restriction onBinding Middleware Products to Operat-ing System Products’’ unless Microsoftalso offers consumers ‘‘an otherwise identi-cal version’’ of the operating system with-out the middleware. Id.

B. Trial Proceedings

[58–60] Microsoft’s first contention—that the District Court erred by adoptingan expedited trial schedule and receivingevidence through summary witnesses—iseasily disposed of. Trial courts have ex-traordinarily broad discretion to determinethe manner in which they will conducttrials. ‘‘This is particularly true in a casesuch as the one at bar where the proceed-ings are being tried to the court without ajury.’’ Eli Lilly & Co., Inc. v. GenerixDrug Sales, Inc., 460 F.2d 1096, 1105 (5thCir.1972). In such cases, ‘‘[a]n appellatecourt will not interfere with the trialcourt’s exercise of its discretion to controlits docket and dispatch its business TTT

except upon the clearest showing that theprocedures have resulted in actual andsubstantial prejudice to the complaininglitigant.’’ Id. Microsoft fails to clear thishigh hurdle. Although the companyclaims that setting an early trial date in-hibited its ability to conduct discovery, itnever identified a specific deposition ordocument it was unable to obtain. Andwhile Microsoft now argues that the use ofsummary witnesses made inevitable theimproper introduction of hearsay evidence,the company actually agreed to the Dis-trict Court’s proposal to limit each side to12 summary witnesses. 12/2/98 am Tr. at11, reprinted in 21 J.A. at 14083 (courtadmonishing Microsoft’s counsel to ‘‘[k]eepin mind that both sides agreed to thenumber of witnesses’’). Even absent Mi-crosoft’s agreement, the company’s chal-lenge fails to show that this use of sum-


mary witnesses falls outside the trialcourt’s wide latitude to receive evidence asit sees fit. General Elec. Co. v. Joiner, 522U.S. 136, 141–42, 118 S.Ct. 512, 139L.Ed.2d 508 (1997). This is particularlytrue given the presumption that a judgewho conducts a bench trial has ignored anyinadmissible evidence, Harris v. Rivera,454 U.S. 339, 346, 102 S.Ct. 460, 70L.Ed.2d 530 (1981)—a presumption thatMicrosoft makes no serious attempt toovercome. Indeed, under appropriate cir-cumstances with appropriate instructions,we have in the past approved the use ofsummary witnesses even in jury trials.See, e.g., United States v. Lemire, 720 F.2d1327 (D.C.Cir.1983). Therefore, neitherthe use of the summary witnesses nor anyother aspect of the District Court’s con-duct of the trial phase amounted to anabuse of discretion.

C. Failure to Hold an Evidentiary Hear-ing

[61, 62] The District Court’s remedies-phase proceedings are a different matter.It is a cardinal principle of our system ofjustice that factual disputes must be heardin open court and resolved through trial-like evidentiary proceedings. Any othercourse would be contrary ‘‘to the spiritwhich imbues our judicial tribunals prohib-iting decision without hearing.’’ Sims v.Greene, 161 F.2d 87, 88 (3d Cir.1947).

[63, 64] A party has the right to judi-cial resolution of disputed facts not just asto the liability phase, but also as to appro-priate relief. ‘‘Normally, an evidentiaryhearing is required before an injunctionmay be granted.’’ United States v.McGee, 714 F.2d 607, 613 (6th Cir.1983);see also Charlton v. Estate of Charlton,841 F.2d 988, 989 (9th Cir.1988) (‘‘General-ly the entry or continuation of an injunc-tion requires a hearing. Only when thefacts are not in dispute, or when the ad-

verse party has waived its right to a hear-ing, can that significant procedural step beeliminated.’’ (citation and internal quota-tion marks omitted)). Other than a tem-porary restraining order, no injunctive re-lief may be entered without a hearing.See generally FED. R. CIV. P. 65. A hear-ing on the merits—i.e., a trial on liability—does not substitute for a relief-specific evi-dentiary hearing unless the matter of re-lief was part of the trial on liability, orunless there are no disputed factual issuesregarding the matter of relief.

[65] This rule is no less applicable inantitrust cases. The Supreme Court ‘‘hasrecognized that a ‘full exploration of factsis usually necessary in order (for the Dis-trict Court) properly to draw (an antitrust)decree’ so as ‘to prevent future violationsand eradicate existing evils.’ ’’ UnitedStates v. Ward Baking Co., 376 U.S. 327,330–31, 84 S.Ct. 763, 11 L.Ed.2d 743 (1964)(quoting Associated Press v. UnitedStates, 326 U.S. 1, 22, 65 S.Ct. 1416, 89L.Ed. 2013 (1945)). Hence a remedies de-cree must be vacated whenever there is ‘‘abona fide disagreement concerning sub-stantive items of relief which could beresolved only by trial.’’ Id. at 334, 84 S.Ct.763; cf. Sims, 161 F.2d at 89 (‘‘It hasnever been supposed that a temporary in-junction could issue under the Clayton Actwithout giving the party against whom theinjunction was sought an opportunity topresent evidence on his behalf.’’).

Despite plaintiffs’ protestations, therecan be no serious doubt that the partiesdisputed a number of facts during theremedies phase. In two separate offers ofproof, Microsoft identified 23 witnesseswho, had they been permitted to testify,would have challenged a wide range ofplaintiffs’ factual representations, includingthe feasibility of dividing Microsoft, thelikely impact on consumers, and the effectof divestiture on shareholders. To take


but two examples, where plaintiffs’ econo-mists testified that splitting Microsoft intwo would be socially beneficial, the com-pany offered to prove that the proposedremedy would ‘‘cause substantial socialharm by raising software prices, loweringrates of innovation and disrupting the evo-lution of Windows as a software develop-ment platform.’’ Defendant’s Offer ofProof at 6, reprinted in 4 J.A. at 2747.And where plaintiffs’ investment bankingexperts proposed that divestiture mightactually increase shareholder value, Micro-soft proffered evidence that structural re-lief ‘‘would inevitably result in a significantloss of shareholder value,’’ a loss that couldreach ‘‘tens—possibly hundreds—of bil-lions of dollars.’’ Id. at 19, reprinted in 4J.A. at 2760.

Indeed, the District Court itself appearsto have conceded the existence of acutefactual disagreements between Microsoftand plaintiffs. The court acknowledgedthat the parties were ‘‘sharply divided’’and held ‘‘divergent opinions’’ on the likelyresults of its remedies decree. FinalJudgment, at 62. The reason the courtdeclined to conduct an evidentiary hearingwas not because of the absence of disputedfacts, but because it believed that thosedisputes could be resolved only through‘‘actual experience,’’ not further proceed-ings. Id. But a prediction about futureevents is not, as a prediction, any less afactual issue. Indeed, the Supreme Courthas acknowledged that drafting an anti-trust decree by necessity ‘‘involves predic-tions and assumptions concerning futureeconomic and business events.’’ Ford Mo-tor Co. v. United States, 405 U.S. 562, 578,92 S.Ct. 1142, 31 L.Ed.2d 492 (1972). Tri-al courts are not excused from their obli-gation to resolve such matters throughevidentiary hearings simply because theyconsider the bedrock procedures of ourjustice system to be ‘‘of little use.’’ FinalJudgment, at 62.

The presence of factual disputes thusdistinguishes this case from the decisionsplaintiffs cite for the proposition that Mi-crosoft was not entitled to an evidentiaryhearing. Indeed, far from assisting plain-tiffs, these cases actually confirm the prop-osition that courts must hold evidentiaryhearings when they are confronted withdisputed facts. In Ford Motor Co., theSupreme Court affirmed a divestiture or-der after emphasizing that the DistrictCourt had ‘‘held nine days of hearings onthe remedy.’’ 405 U.S. at 571, 92 S.Ct.1142. In Davoll v. Webb, 194 F.3d 1116(10th Cir.1999), the defendant both failedto submit any offers of proof, and waivedits right to an evidentiary hearing by ex-pressly agreeing that relief should be de-termined based solely on written submis-sions. Id. at 1142–43. The defendants inAmerican Can Co. v. Mansukhani, 814F.2d 421 (7th Cir.1987), were not entitledto a hearing on remedies because theyfailed ‘‘to explain to the district court whatnew proof they would present to show’’that the proposed remedy was unwarrant-ed. Id. at 425. And in Socialist WorkersParty v. Illinois State Board of Elections,566 F.2d 586 (7th Cir.1977), aff’d, 440 U.S.173, 99 S.Ct. 983, 59 L.Ed.2d 230 (1979),the Seventh Circuit held that a remedies-specific hearing was unnecessary becausethat case involved a pure question of legalinterpretation and hence ‘‘[t]here was nofactual dispute as to the ground on whichthe injunction was ordered.’’ Id. at 587.

Unlike the parties in Davoll, AmericanCan, and Socialist Workers Party, Micro-soft both repeatedly asserted its right toan evidentiary hearing and submitted twooffers of proof. The company’s ‘‘summaryresponse’’ to the proposed remedy arguedthat it would be ‘‘impossible’’ to addressunderlying factual issues ‘‘on a highly ex-pedited basis,’’ Defendant’s Summary Re-sponse at 6–7, reprinted in 4 J.A. at 2587–


88, and Microsoft further maintained thatthe court could not issue a decree unless itfirst permitted ‘‘substantial discovery, ade-quate time for preparation and a full trialon relief.’’ Defendant’s Position as to Fu-ture Proceedings at 2, reprinted in 4 J.A.at 2646. And in 53 pages of submissions,Microsoft identified the specific evidence itwould introduce to challenge plaintiffs’representations.

[66] Plaintiffs further argue—and theDistrict Court held—that no evidentiaryhearing was necessary given that Micro-soft long had been on notice that structur-al relief was a distinct possibility. It isdifficult to see why this matters. WhetherMicrosoft had advance notice that dissolu-tion was in the works is immaterial towhether the District Court violated thecompany’s procedural rights by ordering itwithout an evidentiary hearing. To besure, ‘‘claimed surprise at the districtcourt’s decision to consider permanent in-junctive relief does not, alone, merit rever-sal.’’ Socialist Workers, 566 F.2d at 587.But in this case, Microsoft’s professed sur-prise does not stand ‘‘alone.’’ There issomething more: the company’s basic pro-cedural right to have disputed facts re-solved through an evidentiary hearing.

In sum, the District Court erred when itresolved the parties’ remedies-phase factu-al disputes by consulting only the evidenceintroduced during trial and plaintiffs’ rem-edies phase submissions, without consider-ing the evidence Microsoft sought to intro-duce. We therefore vacate the DistrictCourt’s final judgment, and remand withinstructions to conduct a remedies-specificevidentiary hearing.

D. Failure to Provide an Adequate Ex-planation

[67] We vacate the District Court’sremedies decree for the additional reasonthat the court has failed to provide an

adequate explanation for the relief it or-dered. The Supreme Court has explainedthat a remedies decree in an antitrust casemust seek to ‘‘unfetter a market from anti-competitive conduct,’’ Ford Motor Co., 405U.S. at 577, 92 S.Ct. 1142, to ‘‘terminatethe illegal monopoly, deny to the defendantthe fruits of its statutory violation, andensure that there remain no practices like-ly to result in monopolization in the fu-ture,’’ United States v. United Shoe Mach.Corp., 391 U.S. 244, 250, 88 S.Ct. 1496, 20L.Ed.2d 562 (1968); see also United Statesv. Grinnell Corp., 384 U.S. 563, 577, 86S.Ct. 1698, 16 L.Ed.2d 778 (1966).

The District Court has not explainedhow its remedies decree would accomplishthose objectives. Indeed, the court devot-ed a mere four paragraphs of its order toexplaining its reasons for the remedy.They are: (1) Microsoft ‘‘does not yet con-cede that any of its business practicesviolated the Sherman Act’’; (2) Microsoft‘‘continues to do business as it has in thepast’’; (3) Microsoft ‘‘has proved untrust-worthy in the past’’; and (4) the Govern-ment, whose officials ‘‘are by reason ofoffice obliged and expected to consider—and to act in—the public interest,’’ won thecase, ‘‘and for that reason alone have someentitlement to a remedy of their choice.’’Final Judgment, at 62–63. Nowhere didthe District Court discuss the objectivesthe Supreme Court deems relevant.

E. Modification of Liability

[68] Quite apart from its proceduraldifficulties, we vacate the District Court’sfinal judgment in its entirety for the addi-tional, independent reason that we havemodified the underlying bases of liability.Of the three antitrust violations originallyidentified by the District Court, one is nolonger viable: attempted monopolization ofthe browser market in violation of Sher-man Act § 2. One will be remanded for


liability proceedings under a different legalstandard: unlawful tying in violation of§ 1. Only liability for the § 2 monopoly-maintenance violation has been affirmed—and even that we have revised. Ordinari-ly, of course, we review the grant or denialof equitable relief under the abuse of dis-cretion standard. See, e.g., Doran v. Sa-lem Inn, Inc., 422 U.S. 922, 931–32, 95S.Ct. 2561, 45 L.Ed.2d 648 (1975) (‘‘[T]hestandard of appellate review is simplywhether the issuance of the injunction, inthe light of the applicable standard, consti-tuted an abuse of discretion.’’). For obvi-ous reasons, the application of that stan-dard is not sufficient to sustain the remedyin the case before us. We cannot determinewhether the District Court has abused itsdiscretion in remedying a wrong where thecourt did not exercise that discretion inorder to remedy the properly determinedwrong. That is, the District Court deter-mined that the conduct restrictions andthe pervasive structural remedy were to-gether appropriate to remedy the threeantitrust violations set forth above. Thecourt did not exercise its discretion todetermine whether all, or for that matter,any, of those equitable remedies were re-quired to rectify a § 2 monopoly mainte-nance violation taken alone. We thereforecannot sustain an exercise of discretion notyet made.

By way of comparison, in SpectrumSports, Inc. v. McQuillan, 506 U.S. 447,113 S.Ct. 884, 122 L.Ed.2d 247 (1993), theSupreme Court reviewed a damages awardin a Sherman Act case. In that case, thetrial court entered judgment upon a juryverdict which did not differentiate amongmultiple possible theories of liability under§ 2. The Supreme Court ultimately deter-mined that the trial record could not legal-ly support a finding that the defendant hadcommitted an illegal attempt to monopo-lize, and that ‘‘the trial instructions allowedthe jury to infer specific intent and dan-

gerous probability of success from the de-fendants’ predatory conduct, without anyproof of the relevant market or of a realis-tic probability that the defendants couldachieve monopoly power in that market.’’Id. at 459, 113 S.Ct. 884. Therefore, theHigh Court reversed the Ninth Circuit’sjudgment affirming the District Court andremanded for further proceedings, ex-pressly because ‘‘the jury’s verdict did notnegate the possibility that the § 2 verdictrested on the attempt to monopolizegrounds aloneTTTT’’ Id. Similarly, here,we cannot presume that a District Courtwould exercise its discretion to fashion thesame remedy where the erroneousgrounds of liability were stripped from itsconsideration.

The Eighth Circuit confronted a similarproblem in Concord Boat Corp. v. Bruns-wick Corp., 207 F.3d 1039 (8th Cir.), cert.denied, 531 U.S. 979, 121 S.Ct. 428, 148L.Ed.2d 436 (2000). In that case, a groupof boat builders brought an action againstan engine manufacturer alleging violationsof Sherman Act §§ 1 and 2, and ClaytonAct § 7. After a 10–week trial, the juryfound Brunswick liable on all three countsand returned a verdict for over $44 million.On appeal, the Eighth Circuit reversed theClayton Act claim. Id. at 1053. Thatcourt held that, as a consequence, it wasrequired to vacate the jury’s remedy in itsentirety. Because the ‘‘verdict form didnot require the jury to consider what dam-ages resulted from each type of violation,’’the court could not ‘‘know what damages itfound to have been caused by the acquisi-tions upon which the Section 7 claims werebased.’’ Id. at 1054. The court rejectedthe proposition that ‘‘the entire damageaward may be upheld based on Bruns-wick’s Sherman Act liability alone,’’ id. at1053, holding that, because ‘‘there is noway to know what damages the jury as-signed to the Section 7 claims,’’ the defen-


dant ‘‘would be entitled at the very least toa new damages trial on the boat builders’Sherman Act claims,’’ id. at 1054.

[69] Spectrum Sports and ConcordBoat are distinguishable from the case be-fore us in that both involved the award ofmoney damages rather than equitable re-lief. Nonetheless, their reasoning is in-structive. A court in both contexts mustbase its relief on some clear ‘‘indication ofa significant causal connection between theconduct enjoined or mandated and the vio-lation found directed toward the remedialgoal intended.’’ 3 PHILLIP E. AREEDA &HERBERT HOVENKAMP, ANTITRUST LAW

¶ 653(b), at 91–92 (1996). In a case suchas the one before us where sweeping equi-table relief is employed to remedy multipleviolations, and some—indeed most—of thefindings of remediable violations do notwithstand appellate scrutiny, it is neces-sary to vacate the remedy decree since theimplicit findings of causal connection nolonger exist to warrant our deferential af-firmance.

In short, we must vacate the remediesdecree in its entirety and remand the casefor a new determination. This court hasdrastically altered the District Court’s con-clusions on liability. On remand, the Dis-trict Court, after affording the parties aproper opportunity to be heard, can fash-ion an appropriate remedy for Microsoft’santitrust violations. In particular, thecourt should consider which of the decree’sconduct restrictions remain viable in lightof our modification of the original liabilitydecision. While the task of drafting theremedies decree is for the District Courtin the first instance, because of the unusu-ally convoluted nature of the proceedingsthus far, and a desire to advance the ulti-mate resolution of this important contro-versy, we offer some further guidance forthe exercise of that discretion.

F. On Remand

[70] As a general matter, a districtcourt is afforded broad discretion to enterthat relief it calculates will best remedythe conduct it has found to be unlawful.See, e.g., Woerner v. United States SmallBus. Admin., 934 F.2d 1277, 1279(D.C.Cir.1991) (recognizing that an appel-late court reviews a trial court’s decisionwhether or not to grant equitable reliefonly for an abuse of discretion). This is noless true in antitrust cases. See, e.g., FordMotor Co., 405 U.S. at 573, 92 S.Ct. 1142(‘‘The District Court is clothed with ‘largediscretion’ to fit the decree to the specialneeds of the individual case.’’); Md. & Va.Milk Producers Ass’n, Inc. v. UnitedStates, 362 U.S. 458, 473, 80 S.Ct. 847, 4L.Ed.2d 880 (1960) (‘‘The formulation ofdecrees is largely left to the discretion ofthe trial courtTTTT’’). And divestiture is acommon form of relief in successful anti-trust prosecutions: it is indeed ‘‘the mostimportant of antitrust remedies.’’ See,e.g., United States v. E.I. du Pont deNemours & Co., 366 U.S. 316, 331, 81 S.Ct.1243, 6 L.Ed.2d 318 (1961).

On remand, the District Court must re-consider whether the use of the structuralremedy of divestiture is appropriate withrespect to Microsoft, which argues that itis a unitary company. By and large, casesupon which plaintiffs rely in arguing forthe split of Microsoft have involved thedissolution of entities formed by mergersand acquisitions. On the contrary, theSupreme Court has clarified that divesti-ture ‘‘has traditionally been the remedyfor Sherman Act violations whose heart isintercorporate combination and control,’’du Pont, 366 U.S. at 329, 81 S.Ct. 1243(emphasis added), and that ‘‘[c]omplete di-vestiture is particularly appropriate whereasset or stock acquisitions violate the anti-trust laws,’’ Ford Motor Co., 405 U.S. at573, 92 S.Ct. 1142 (emphasis added).


One apparent reason why courts havenot ordered the dissolution of unitary com-panies is logistical difficulty. As the courtexplained in United States v. ALCOA, 91F.Supp. 333, 416 (S.D.N.Y.1950), a ‘‘corpo-ration, designed to operate effectively as asingle entity, cannot readily be dismem-bered of parts of its various operationswithout a marked loss of efficiency.’’ Acorporation that has expanded by acquir-ing its competitors often has preexistinginternal lines of division along which itmay more easily be split than a corpora-tion that has expanded from naturalgrowth. Although time and corporatemodifications and developments may even-tually fade those lines, at least the identifi-able entities preexisted to create a temp-late for such division as the court mightlater decree. With reference to those cor-porations that are not acquired by mergerand acquisition, Judge Wyzanski accurate-ly opined in United Shoe:

United conducts all machine manufac-ture at one plant in Beverly, with oneset of jigs and tools, one foundry, onelaboratory for machinery problems, onemanagerial staff, and one labor force. Ittakes no Solomon to see that this organ-ism cannot be cut into three equal andviable parts.

United States v. United Shoe MachineCorp., 110 F.Supp. 295, 348 (D.Mass.1953).

Depending upon the evidence, the Dis-trict Court may find in a remedies pro-ceeding that it would be no easier to splitMicrosoft in two than United Shoe inthree. Microsoft’s Offer of Proof in re-sponse to the court’s denial of an eviden-tiary hearing included proffered testimonyfrom its President and CEO Steve Ballmerthat the company ‘‘is, and always has been,a unified company without free-standingbusiness units. Microsoft is not the resultof mergers or acquisitions.’’ Microsoftfurther offered evidence that it is ‘‘not

organized along product lines,’’ but ratheris housed in a single corporate headquar-ters and that it has

only one sales and marketing organiza-tion which is responsible for selling all ofthe company’s products, one basic re-search organization, one product supportorganization, one operations department,one information technology department,one facilities department, one purchas-ing department, one human resourcesdepartment, one finance department,one legal department and one public re-lations department.

Defendant’s Offer of Proof at 23–26, re-printed in 4 J.A. at 2764–67. If indeedMicrosoft is a unitary company, divisionmight very well require Microsoft to re-produce each of these departments in eachnew entity rather than simply allocate thediffering departments among them.

In devising an appropriate remedy, theDistrict Court also should consider wheth-er plaintiffs have established a sufficientcausal connection between Microsoft’s anti-competitive conduct and its dominant posi-tion in the OS market. ‘‘Mere existence ofan exclusionary act does not itself justifyfull feasible relief against the monopolist tocreate maximum competition.’’ 3 AREEDA

& HOVENKAMP, ANTITRUST LAW ¶ 650a, at 67.Rather, structural relief, which is ‘‘de-signed to eliminate the monopoly altogeth-er TTT require[s] a clearer indication of asignificant causal connection between theconduct and creation or maintenance of themarket power.’’ Id. ¶ 653b, at 91–92 (em-phasis added). Absent such causation, theantitrust defendant’s unlawful behaviorshould be remedied by ‘‘an injunctionagainst continuation of that conduct.’’ Id.¶ 650a, at 67.

As noted above, see supra Section II.C,we have found a causal connection betweenMicrosoft’s exclusionary conduct and itscontinuing position in the operating sys-


tems market only through inference. See3 AREEDA & HOVENKAMP, ANTITRUST LAW

¶ 653(b), at 91–92 (suggesting that ‘‘moreextensive equitable relief, particularly rem-edies such as divestiture designed to elimi-nate the monopoly altogether, TTT requirea clearer indication of significant causalconnection between the conduct and cre-ation or maintenance of the market pow-er’’). Indeed, the District Court expresslydid not adopt the position that Microsoftwould have lost its position in the OSmarket but for its anticompetitive behav-ior. Findings of Fact ¶ 411 (‘‘There isinsufficient evidence to find that, absentMicrosoft’s actions, Navigator and Java al-ready would have ignited genuine competi-tion in the market for Intel-compatible PCoperating systems.’’). If the court on re-mand is unconvinced of the causal connec-tion between Microsoft’s exclusionary con-duct and the company’s position in the OSmarket, it may well conclude that divesti-ture is not an appropriate remedy.

While we do not undertake to dictate tothe District Court the precise form thatrelief should take on remand, we noteagain that it should be tailored to fit thewrong creating the occasion for the reme-dy.

G. Conclusion

In sum, we vacate the District Court’sremedies decree for three reasons. First,the District Court failed to hold an eviden-tiary hearing despite the presence of rem-edies-specific factual disputes. Second,the court did not provide adequate reasonsfor its decreed remedies. Finally, we havedrastically altered the scope of Microsoft’sliability, and it is for the District Court inthe first instance to determine the propri-ety of a specific remedy for the limitedground of liability which we have upheld.

VI. JUDICIAL MISCONDUCT

Canon 3A(6) of the Code of Conduct forUnited States Judges requires federaljudges to ‘‘avoid public comment on themerits of [ ] pending or impending’’ cases.Canon 2 tells judges to ‘‘avoid improprietyand the appearance of impropriety in allactivities,’’ on the bench and off. Canon3A(4) forbids judges to initiate or considerex parte communications on the merits ofpending or impending proceedings. Sec-tion 455(a) of the Judicial Code requiresjudges to recuse themselves when their‘‘impartiality might reasonably be ques-tioned.’’ 28 U.S.C. § 455(a).

All indications are that the DistrictJudge violated each of these ethical pre-cepts by talking about the case with re-porters. The violations were deliberate,repeated, egregious, and flagrant. Theonly serious question is what consequencesshould follow. Microsoft urges us to dis-qualify the District Judge, vacate the judg-ment in its entirety and toss out the find-ings of fact, and remand for a new trialbefore a different District Judge. At theother extreme, plaintiffs ask us to do noth-ing. We agree with neither position.

A. The District Judge’s Communicationswith the Press

Immediately after the District Judgeentered final judgment on June 7, 2000,accounts of interviews with him began ap-pearing in the press. Some of the inter-views were held after he entered finaljudgment. See Peter Spiegel, MicrosoftJudge Defends Post-trial Comments, FIN.

TIMES (London), Oct. 7, 2000, at 4; John R.Wilke, For Antitrust Judge, Trust, orLack of It, Really Was the Issue—In anInterview, Jackson Says Microsoft Didthe Damage to Its Credibility in Court,WALL ST. J., June 8, 2000, at A1. TheDistrict Judge also aired his views aboutthe case to larger audiences, giving


speeches at a college and at an antitrustseminar. See James V. Grimaldi, Micro-soft Judge Says Ruling at Risk; EveryTrial Decision Called ‘Vulnerable’, WASH.

POST, Sept. 29, 2000, at E1; AlisonSchmauch, Microsoft Judge Shares Expe-riences, THE DARTMOUTH ONLINE, Oct. 3,2000.

From the published accounts, it is ap-parent that the Judge also had been givingsecret interviews to select reporters beforeentering final judgment—in some instanc-es long before. The earliest interviews weknow of began in September 1999, shortlyafter the parties finished presenting evi-dence but two months before the courtissued its Findings of Fact. See JoelBrinkley & Steve Lohr, U.S. vs. Microsoft:Pursuing a Giant; Retracing the Misstepsin the Microsoft Defense, N.Y. TIMES, June9, 2000, at A1. Interviews with reportersfrom the New York Times and Ken Aulet-ta, another reporter who later wrote abook on the Microsoft case, continuedthroughout late 1999 and the first half of2000, during which time the Judge issuedhis Findings of Fact, Conclusions of Law,and Final Judgment. See id.; Ken Aulet-ta, Final Offer, THE NEW YORKER, Jan. 15,2001, at 40. The Judge ‘‘embargoed’’these interviews; that is, he insisted thatthe fact and content of the interviews re-main secret until he issued the Final Judg-ment.

[71] Before we recount the statementsattributed to the District Judge, we needto say a few words about the state of therecord. All we have are the publishedaccounts and what the reporters say theJudge said. Those accounts were not ad-mitted in evidence. They may be hearsay.See FED.R.EVID. 801(c); Metro. Council ofNAACP Branches v. FCC, 46 F.3d 1154,1165 (D.C.Cir.1995) (‘‘We seriously ques-tion whether a New York Times article is

admissible evidence of the truthfulness ofits contents.’’).

We are of course concerned about grant-ing a request to disqualify a federal judgewhen the material supporting it has notbeen admitted in evidence. Disqualifica-tion is never taken lightly. In the wronghands, a disqualification motion is a proce-dural weapon to harass opponents and de-lay proceedings. If supported only by ru-mor, speculation, or innuendo, it is also ameans to tarnish the reputation of a feder-al judge.

But the circumstances of this case aremost unusual. By placing an embargo onthe interviews, the District Judge ensuredthat the full extent of his actions would notbe revealed until this case was on appeal.Plaintiffs, in defending the judgment, donot dispute the statements attributed tohim in the press; they do not request anevidentiary hearing; and they do not ar-gue that Microsoft should have filed amotion in the District Court before raisingthe matter on appeal. At oral argument,plaintiffs all but conceded that the Judgeviolated ethical restrictions by discussingthe case in public: ‘‘On behalf of the gov-ernments, I have no brief to defend theDistrict Judge’s decision to discuss thiscase publicly while it was pending on ap-peal, and I have no brief to defend thejudge’s decision to discuss the case withreporters while the trial was proceeding,even given the embargo on any reportingconcerning those conversations until afterthe trial.’’ 02/27/01 Ct. Appeals Tr. at 326.

[72] We must consider too that thefederal disqualification provisions reflect astrong federal policy to preserve the actu-al and apparent impartiality of the federaljudiciary. Judicial misconduct may impli-cate that policy regardless of the meansby which it is disclosed to the public. Cf.The Washington Post v. Robinson, 935F.2d 282, 291 (D.C.Cir.1991) (taking judi-


cial notice of newspaper articles to ascer-tain whether a fact was within publicknowledge). Also, in our analysis of thearguments presented by the parties, thespecifics of particular conversations areless important than their cumulative ef-fect.

[73] For these reasons we have decid-ed to adjudicate Microsoft’s disqualifica-tion request notwithstanding the state ofthe record. The same reasons also war-rant a departure from our usual practice ofdeclining to address issues raised for thefirst time on appeal: the ‘‘matter of whatquestions may be taken up and resolvedfor the first time on appeal is one leftprimarily to the discretion of the courts ofappeals, to be exercised on the facts ofindividual cases.’’ Singleton v. Wulff, 428U.S. 106, 121, 96 S.Ct. 2868, 49 L.Ed.2d826 (1976); accord Hormel v. Helvering,312 U.S. 552, 556–57, 61 S.Ct. 719, 85L.Ed. 1037 (1941); Nat’l Ass’n of Mfrs. v.Dep’t of Labor, 159 F.3d 597, 605–06(D.C.Cir.1998). We will assume the truthof the press accounts and not send thecase back for an evidentiary hearing onthis subject. We reach no judgment onwhether the details of the interviews wereaccurately recounted.

The published accounts indicate that theDistrict Judge discussed numerous topicsrelating to the case. Among them was hisdistaste for the defense of technologicalintegration—one of the central issues inthe lawsuit. In September 1999, twomonths before his Findings of Fact and sixmonths before his Conclusions of Law, andin remarks that were kept secret untilafter the Final Judgment, the Judge toldreporters from the New York Times thathe questioned Microsoft’s integration of aweb browser into Windows. Stating thathe was ‘‘not a fan of integration,’’ he drewan analogy to a 35–millimeter camera withan integrated light meter that in his view

should also be offered separately: ‘‘Youlike the convenience of having a light me-ter built in, integrated, so all you have todo is press a button to get a reading. Butdo you think camera makers should alsoserve photographers who want to use aseparate light meter, so they can hold itup, move it around?’’ JOEL BRINKLEY &STEVE LOHR, U.S. V. MICROSOFT 263 (2001).In other remarks, the Judge commentedon the integration at the heart of the case:‘‘[I]t was quite clear to me that the motiveof Microsoft in bundling the Internetbrowser was not one of consumer conve-nience. The evidence that this was donefor the consumer was not credibleTTTT

The evidence was so compelling that therewas an ulterior motive.’’ Wilke, WALL ST.

J. As for tying law in general, he criti-cized this court’s ruling in the consentdecree case, saying it ‘‘was wrongheadedon several counts’’ and would exempt thesoftware industry from the antitrust laws.BRINKLEY & LOHR, U.S. V. MICROSOFT 78,295; Brinkley & Lohr, N.Y. TIMES.

Reports of the interviews have the Dis-trict Judge describing Microsoft’s conduct,with particular emphasis on what he re-garded as the company’s prevarication, hu-bris, and impenitence. In some of hissecret meetings with reporters, the Judgeoffered his contemporaneous impressionsof testimony. He permitted at least onereporter to see an entry concerning BillGates in his ‘‘oversized green notebook.’’KEN AULETTA, WORLD WAR 3.0, at 112(2001). He also provided numerous after-the-fact credibility assessments. He toldreporters that Bill Gates’ ‘‘testimony isinherently without credibility’’ and ‘‘[i]f youcan’t believe this guy, who else can youbelieve?’’ BRINKLEY & LOHR, U.S. V. MICRO-

SOFT 278; Brinkley & Lohr, N.Y. TIMES;see also Auletta, THE NEW YORKER, at 40.As for the company’s other witnesses, theJudge is reported as saying that there


‘‘were times when I became impatient withMicrosoft witnesses who were givingspeeches.’’ ‘‘[T]hey were telling me thingsI just flatly could not credit.’’ Brinkley &Lohr, N.Y. TIMES. In an interview giventhe day he entered the break-up order, hesummed things up: ‘‘Falsus in uno, falsusin omnibus’’: ‘‘Untrue in one thing, untruein everything.’’ ‘‘I don’t subscribe to thatas absolutely true. But it does lead one tosuspicion. It’s a universal human experi-ence. If someone lies to you once, howmuch else can you credit as the truth?’’Wilke, WALL ST. J.

According to reporter Auletta, the Dis-trict Judge told him in private that, ‘‘Ithought they [Microsoft and its executives]didn’t think they were regarded as adultmembers of the community. I thoughtthey would learn.’’ AULETTA, WORLD WAR

3.0, at 14. The Judge told a college audi-ence that ‘‘Bill Gates is an ingenious engi-neer, but I don’t think he is that adept atbusiness ethics. He has not yet come torealise things he did (when Microsoft wassmaller) he should not have done when hebecame a monopoly.’’ Spiegel, FIN. TIMES.Characterizing Gates’ and his company’s‘‘crime’’ as hubris, the Judge stated that‘‘[i]f I were able to propose a remedy ofmy devising, I’d require Mr. Gates to writea book report’’ on Napoleon Bonaparte,‘‘[b]ecause I think [Gates] has a Napoleon-ic concept of himself and his company, anarrogance that derives from power andunalloyed success, with no leavening hardexperience, no reverses.’’ Auletta, THE

NEW YORKER, at 41; see also AULETTA,

WORLD WAR 3.0, at 397. The Judge appar-ently became, in Auletta’s words, ‘‘increas-ingly troubled by what he learned aboutBill Gates and couldn’t get out of his mindthe group picture he had seen of Bill Gatesand Paul Allen and their shaggy-hairedfirst employees at Microsoft.’’ The report-er wrote that the Judge said he saw in thepicture ‘‘a smart-mouthed young kid who

has extraordinary ability and needs a littlediscipline. I’ve often said to colleaguesthat Gates would be better off if he hadfinished Harvard.’’ AULETTA, WORLD WAR

3.0, at 168–69; see also Auletta, THE NEW

YORKER, at 46 (reporting the DistrictJudge’s statement that ‘‘they [Microsoftand its executives] don’t act like grown-ups!’’ ‘‘[T]o this day they continue to denythey did anything wrong.’’).

The District Judge likened Microsoft’swriting of incriminating documents to drugtraffickers who ‘‘never figure out that theyshouldn’t be saying certain things on thephone.’’ BRINKLEY & LOHR, U.S. V. MICRO-

SOFT 6; Brinkley & Lohr, N.Y. TIMES. Heinvoked the drug trafficker analogy againto denounce Microsoft’s protestations ofinnocence, this time with a reference to thenotorious Newton Street Crew that terror-ized parts of Washington, D.C. ReporterAuletta wrote in The New Yorker that theJudge

went as far as to compare the company’sdeclaration of innocence to the protesta-tions of gangland killers. He was refer-ring to five gang members in a racke-teering, drug-dealing, and murder trialthat he had presided over four yearsearlier. In that case, the three victimshad had their heads bound with ducttape before they were riddled with bul-lets from semi-automatic weapons. ‘‘Onthe day of the sentencing, the gangmembers maintained that they had donenothing wrong, saying that the wholecase was a conspiracy by the white pow-er structure to destroy them,’’ Jacksonrecalled. ‘‘I am now under no illusionsthat miscreants will realize that otherparts of society will view them thatway.’’

Auletta, THE NEW YORKER, at 40–41; AU-

LETTA, WORLD WAR 3.0, at 369–70 (same);see also Auletta, THE NEW YORKER, at 46.


The District Judge also secretly di-vulged to reporters his views on the reme-dy for Microsoft’s antitrust violations. Onthe question whether Microsoft was enti-tled to any process at the remedy stage,the Judge told reporters in May 2000 thathe was ‘‘not aware of any case authoritythat says I have to give them any dueprocess at all. The case is over. Theylost.’’ Brinkley & Lohr, N.Y. TIMES. An-other reporter has the Judge asking‘‘[w]ere the Japanese allowed to proposeterms of their surrender?’’ Spiegel, FIN.

TIMES. The District Judge also told re-porters the month before he issued hisbreak-up order that ‘‘[a]ssuming, as Ithink they are, [ ] the Justice Departmentand the states are genuinely concernedabout the public interest,’’ ‘‘I know theyhave carefully studied all the possible op-tions. This isn’t a bunch of amateurs.They have consulted with some of the bestminds in America over a long period oftime.’’ ‘‘I am not in a position to duplicatethat and re-engineer their work. There’sno way I can equip myself to do a betterjob than they have done.’’ Brinkley &Lohr, N.Y. TIMES; cf. Final Judgment, at62–63.

In February 2000, four months beforehis final order splitting the company intwo, the District Judge reportedly toldNew York Times reporters that he was‘‘not at all comfortable with restructuringthe company,’’ because he was unsurewhether he was ‘‘competent to do that.’’Brinkley & Lohr, N.Y. TIMES; see alsoBRINKLEY & LOHR, U.S. V. MICROSOFT 277–78 (same); cf. AULETTA, WORLD WAR 3.0, at370 (comment by the Judge in April 2000that he was inclining toward behavioralrather than structural remedies). A fewmonths later, he had a change of heart.He told the same reporters that ‘‘withwhat looks like Microsoft intransigence, abreakup is inevitable.’’ Brinkley & Lohr,N.Y. TIMES; see also BRINKLEY & LOHR,

U.S. V. MICROSOFT 315. The Judge reciteda ‘‘North Carolina mule trainer’’ story toexplain his change in thinking from ‘‘[i]f itain’t broken, don’t try to fix it’’ and ‘‘I justdon’t think that [restructuring the compa-ny] is something I want to try to do on myown’’ to ordering Microsoft broken in two:

He had a trained mule who could do allkinds of wonderful tricks. One daysomebody asked him: ‘‘How do you doit? How do you train the mule to do allthese amazing things?’’ ‘‘Well,’’ he an-swered, ‘‘I’ll show you.’’ He took a 2–by–4 and whopped him upside the head.The mule was reeling and fell to hisknees, and the trainer said: ‘‘You justhave to get his attention.’’

BRINKLEY & LOHR, U.S. V. MICROSOFT 278.The Judge added: ‘‘I hope I’ve got Micro-soft’s attention.’’ Id.; see also Grimaldi,WASH. POST (comments by the Judge blam-ing the break-up on Microsoft’s intransi-gence and on what he perceived to beMicrosoft’s responsibility for the failure ofsettlement talks); Spiegel, FIN. TIMES (theJudge blaming break-up on Microsoft’s in-transigence).

B. Violations of the Code of Conduct forUnited States Judges

[74] The Code of Conduct for UnitedStates Judges was adopted by the JudicialConference of the United States in 1973.It prescribes ethical norms for federaljudges as a means to preserve the actualand apparent integrity of the federal judi-ciary. Every federal judge receives a copyof the Code, the Commentary to the Code,the Advisory Opinions of the Judicial Con-ference’s Committee on Codes of Conduct,and digests of the Committee’s informal,unpublished opinions. See II GUIDE TO

JUDICIARY POLICIES AND PROCEDURES (1973).The material is periodically updated.Judges who have questions about whethertheir conduct would be consistent with the


Code may write to the Codes of ConductCommittee for a written, confidential opin-ion. See Introduction, CODE OF CONDUCT.The Committee traditionally respondspromptly. A judge may also seek informaladvice from the Committee’s circuit repre-sentative.

While some of the Code’s Canons fre-quently generate questions about their ap-plication, others are straightforward andeasily understood. Canon 3A(6) is an ex-ample of the latter. In forbidding federaljudges to comment publicly ‘‘on the meritsof a pending or impending action,’’ Canon3A(6) applies to cases pending before anycourt, state or federal, trial or appellate.See JEFFREY M. SHAMAN ET AL., JUDICIAL

CONDUCT AND ETHICS § 10.34, at 353 (3ded.2000). As ‘‘impending’’ indicates, theprohibition begins even before a case en-ters the court system, when there is rea-son to believe a case may be filed. Cf. E.WAYNE THODE, REPORTER’S NOTES TO CODE OF

JUDICIAL CONDUCT 54 (1973). An actionremains ‘‘pending’’ until ‘‘completion of theappellate process.’’ CODE OF CONDUCT Can-on 3A(6) cmt.; Comm. on Codes of Con-duct, Adv. Op. No. 55 (1998).

The Microsoft case was ‘‘pending’’ dur-ing every one of the District Judge’s meet-ings with reporters; the case is ‘‘pending’’now; and even after our decision issues, itwill remain pending for some time. TheDistrict Judge breached his ethical dutyunder Canon 3A(6) each time he spoke to areporter about the merits of the case. Al-though the reporters interviewed him inprivate, his comments were public. Courtwas not in session and his discussion of thecase took place outside the presence of theparties. He provided his views not tocourt personnel assisting him in the case,but to members of the public. And thesewere not just any members of the public.Because he was talking to reporters, the

Judge knew his comments would eventual-ly receive widespread dissemination.

It is clear that the District Judge wasnot discussing purely procedural matters,which are a permissible subject of publiccomment under one of the Canon’s threenarrowly drawn exceptions. He disclosedhis views on the factual and legal mattersat the heart of the case. His opinionsabout the credibility of witnesses, the va-lidity of legal theories, the culpability ofthe defendant, the choice of remedy, andso forth all dealt with the merits of theaction. It is no excuse that the Judge mayhave intended to ‘‘educate’’ the publicabout the case or to rebut ‘‘public misper-ceptions’’ purportedly caused by the par-ties. See Grimaldi, WASH. POST; MicrosoftJudge Says He May Step down from Caseon Appeal, WALL ST. J., Oct. 30, 2000. Ifthose were his intentions, he could haveaddressed the factual and legal issues ashe saw them—and thought the publicshould see them—in his Findings of Fact,Conclusions of Law, Final Judgment, or ina written opinion. Or he could have heldhis tongue until all appeals were conclud-ed.

Far from mitigating his conduct, theDistrict Judge’s insistence on secrecy—hisembargo—made matters worse. Conceal-ment of the interviews suggests knowledgeof their impropriety. Concealment alsoprevented the parties from nipping his im-proprieties in the bud. Without anyknowledge of the interviews, neither theplaintiffs nor the defendant had a chanceto object or to seek the Judge’s removalbefore he issued his Final Judgment.

Other federal judges have been disquali-fied for making limited public commentsabout cases pending before them. See Inre Boston’s Children First, 244 F.3d 164(1st Cir.2001); In re IBM Corp., 45 F.3d641 (2d Cir.1995); United States v. Cooley,1 F.3d 985 (10th Cir.1993). Given the


extent of the Judge’s transgressions in thiscase, we have little doubt that if the par-ties had discovered his secret liaisons withthe press, he would have been disqualified,voluntarily or by court order. Cf. In reBarry, 946 F.2d 913 (D.C.Cir.1991) (percuriam); id. at 915 (Edwards, J., dissent-ing).

In addition to violating the rule prohibit-ing public comment, the District Judge’sreported conduct raises serious questionsunder Canon 3A(4). That Canon statesthat a ‘‘judge should accord to every per-son who is legally interested in a proceed-ing, or the person’s lawyer, full right to beheard according to law, and, except asauthorized by law, neither initiate nor con-sider ex parte communications on the mer-its, or procedures affecting the merits, of apending or impending proceeding.’’ CODE

OF CONDUCT Canon 3A(4).

What did the reporters convey to theDistrict Judge during their secret ses-sions? By one account, the Judge spent atotal of ten hours giving taped interviewsto one reporter. AULETTA, WORLD WAR 3.0,at 14 n.*. We do not know whether hespent even more time in untaped conversa-tions with the same reporter, nor do weknow how much time he spent with others.But we think it safe to assume that theseinterviews were not monologues. Inter-views often become conversations. Whenreporters pose questions or make asser-tions, they may be furnishing information,information that may reflect their personalviews of the case. The published accountsindicate this happened on at least one oc-casion. Ken Auletta reported, for exam-ple, that he told the Judge ‘‘that Microsoftemployees professed shock that he thoughtthey had violated the law and behavedunethically,’’ at which time the Judge be-came ‘‘agitated’’ by ‘‘Microsoft’s ‘obstina-cy’.’’ Id. at 369. It is clear that Aulettahad views of the case. As he wrote in a

Washington Post editorial, ‘‘[a]nyone whosat in [the District Judge’s] courtroomduring the trial had seen ample evidenceof Microsoft’s sometimes thuggish tactics.’’Ken Auletta, Maligning the MicrosoftJudge, WASH. POST, Mar. 7, 2001, at A23.

The District Judge’s repeated violationsof Canons 3A(6) and 3A(4) also violatedCanon 2, which provides that ‘‘a judgeshould avoid impropriety and the appear-ance of impropriety in all activities.’’ CODE

OF CONDUCT Canon 2; see also In re Chargeof Judicial Misconduct, 47 F.3d 399, 400(10th Cir. Jud. Council 1995) (‘‘The allega-tions of extra-judicial comments cause theCouncil substantial concern under bothCanon 3A(6) and Canon 2 of the JudicialCode of Conduct.’’). Canon 2A requiresfederal judges to ‘‘respect and comply withthe law’’ and to ‘‘act at all times in amanner that promotes public confidence inthe integrity and impartiality of the judi-ciary.’’ CODE OF CONDUCT Canon 2A. TheCode of Conduct is the law with respect tothe ethical obligations of federal judges,and it is clear the District Judge violated iton multiple occasions in this case. Therampant disregard for the judiciary’s ethi-cal obligations that the public witnessed inthis case undoubtedly jeopardizes ‘‘publicconfidence in the integrity’’ of the DistrictCourt proceedings.

Another point needs to be stressed.Rulings in this case have potentially hugefinancial consequences for one of the na-tion’s largest publicly-traded companiesand its investors. The District Judge’ssecret interviews during the trial provideda select few with inside information aboutthe case, information that enabled themand anyone they shared it with to antici-pate rulings before the Judge announcedthem to the world. Although he ‘‘embar-goed’’ his comments, the Judge had no wayof policing the reporters. For all he knewthere may have been trading on the basis


of the information he secretly conveyed.The public cannot be expected to maintainconfidence in the integrity and impartialityof the federal judiciary in the face of suchconduct.

C. Appearance of Partiality

[75] The Code of Conduct contains noenforcement mechanism. See THODE, RE-

PORTER’S NOTES TO CODE OF JUDICIAL CON-

DUCT 43. The Canons, including the onethat requires a judge to disqualify himselfin certain circumstances, see CODE OF CON-

DUCT Canon 3C, are self-enforcing. Thereare, however, remedies extrinsic to theCode. One is an internal disciplinary pro-ceeding, begun with the filing of a com-plaint with the clerk of the court of appealspursuant to 28 U.S.C. § 372(c). Anotheris disqualification of the offending judgeunder either 28 U.S.C. § 144, which re-quires the filing of an affidavit while thecase is in the District Court, or 28 U.S.C.§ 455, which does not. Microsoft urgesthe District Judge’s disqualification under§ 455(a): a judge ‘‘shall disqualify himselfin any proceeding in which his impartialitymight reasonably be questioned.’’ 28U.S.C. § 455(a). The standard for dis-qualification under § 455(a) is an objectiveone. The question is whether a reasonableand informed observer would question thejudge’s impartiality. See In re Barry, 946F.2d at 914; see also In re Aguinda, 241F.3d 194, 201 (2d Cir.2001); RICHARD E.FLAMM, JUDICIAL DISQUALIFICATION § 24.2.1(1996).

‘‘The very purpose of § 455(a) is to pro-mote confidence in the judiciary by avoid-ing even the appearance of improprietywhenever possible.’’ Liljeberg v. HealthServs. Acquisition Corp., 486 U.S. 847,865, 108 S.Ct. 2194, 100 L.Ed.2d 855(1988). As such, violations of the Code ofConduct may give rise to a violation of§ 455(a) if doubt is cast on the integrity of

the judicial process. It has been arguedthat any ‘‘public comment by a judge con-cerning the facts, applicable law, or meritsof a case that is sub judice in his court orany comment concerning the parties ortheir attorneys would raise grave doubtsabout the judge’s objectivity and his will-ingness to reserve judgment until the closeof the proceeding.’’ William G. Ross, Ex-trajudicial Speech: Charting the Bound-aries of Propriety, 2 GEO. J. LEGAL ETHICS

589, 598 (1989). Some courts of appealshave taken a hard line on public com-ments, finding violations of § 455(a) forjudicial commentary on pending cases thatseems mild in comparison to what we areconfronting in this case. See Boston’sChildren First, 244 F.3d 164 (grantingwrit of mandamus ordering district judgeto recuse herself under § 455(a) because ofpublic comments on class certification andstanding in a pending case); In re IBMCorp., 45 F.3d 641 (granting writ of man-damus ordering district judge to recusehimself based in part on the appearance ofpartiality caused by his giving newspaperinterviews); Cooley, 1 F.3d 985 (vacatingconvictions and disqualifying district judgefor appearance of partiality because heappeared on television program Nightlineand stated that abortion protestors in acase before him were breaking the law andthat his injunction would be obeyed).

While § 455(a) is concerned with actualand apparent impropriety, the statute re-quires disqualification only when a judge’s‘‘impartiality might reasonably be ques-tioned.’’ 28 U.S.C. § 455(a). Althoughthis court has condemned public judicialcomments on pending cases, we have notgone so far as to hold that every violationof Canon 3A(6) or every impropriety underthe Code of Conduct inevitably destroysthe appearance of impartiality and thusviolates § 455(a). See In re Barry, 946F.2d at 914; see also Boston’s ChildrenFirst, 244 F.3d at 168; United States v.


Fortier, 242 F.3d 1224, 1229 (10th Cir.2001).

In this case, however, we believe the linehas been crossed. The public commentswere not only improper, but also wouldlead a reasonable, informed observer toquestion the District Judge’s impartiality.Public confidence in the integrity and im-partiality of the judiciary is seriously jeop-ardized when judges secretly share theirthoughts about the merits of pending caseswith the press. Judges who covet publici-ty, or convey the appearance that they do,lead any objective observer to wonderwhether their judgments are being influ-enced by the prospect of favorable cover-age in the media. Discreet and limitedpublic comments may not compromise ajudge’s apparent impartiality, but we havelittle doubt that the District Judge’s con-duct had that effect. Appearance may beall there is, but that is enough to invokethe Canons and § 455(a).

Judge Learned Hand spoke of ‘‘thisAmerica of ours where the passion forpublicity is a disease, and where swarms offoolish, tawdry moths dash with raptureinto its consuming fireTTTT’’ LEARNED

HAND, THE SPIRIT OF LIBERTY 132–33 (2ded.1953). Judges are obligated to resistthis passion. Indulging it compromiseswhat Edmund Burke justly regarded asthe ‘‘cold neutrality of an impartial judge.’’Cold or not, federal judges must maintainthe appearance of impartiality. What wastrue two centuries ago is true today: ‘‘Def-erence to the judgments and rulings ofcourts depends upon public confidence inthe integrity and independence of judges.’’CODE OF CONDUCT Canon 1 cmt. Publicconfidence in judicial impartiality cannotsurvive if judges, in disregard of theirethical obligations, pander to the press.

We recognize that it would be extraordi-nary to disqualify a judge for bias or ap-pearance of partiality when his remarks

arguably reflected what he learned, orwhat he thought he learned, during theproceedings. See Liteky v. United States,510 U.S. 540, 554–55, 114 S.Ct. 1147, 127L.Ed.2d 474 (1994); United States v. Bar-ry, 961 F.2d 260, 263 (D.C.Cir.1992). Butthis ‘‘extrajudicial source’’ rule has nobearing on the case before us. The prob-lem here is not just what the DistrictJudge said, but to whom he said it andwhen. His crude characterizations of Mi-crosoft, his frequent denigrations of BillGates, his mule trainer analogy as a reasonfor his remedy—all of these remarks andothers might not have given rise to a viola-tion of the Canons or of § 455(a) had heuttered them from the bench. See Liteky,510 U.S. at 555–56, 114 S.Ct. 1147; CODE

OF CONDUCT Canon 3A(6) (exception to pro-hibition on public comments for ‘‘state-ments made in the course of the judge’sofficial duties’’). But then Microsoft wouldhave had an opportunity to object, perhapseven to persuade, and the Judge wouldhave made a record for review on appeal.It is an altogether different matter whenthe statements are made outside the court-room, in private meetings unknown to theparties, in anticipation that ultimately theJudge’s remarks would be reported.Rather than manifesting neutrality and im-partiality, the reports of the interviewswith the District Judge convey the impres-sion of a judge posturing for posterity,trying to please the reporters with colorfulanalogies and observations bound to windup in the stories they write. Members ofthe public may reasonably question wheth-er the District Judge’s desire for presscoverage influenced his judgments, indeedwhether a publicity-seeking judge mightconsciously or subconsciously seek thepublicity-maximizing outcome. We be-lieve, therefore, that the District Judge’sinterviews with reporters created an ap-pearance that he was not acting impartial-


ly, as the Code of Conduct and § 455(a)require.

D. Remedies for Judicial Misconductand Appearance of Partiality

1. Disqualification

[76] Disqualification is mandatory forconduct that calls a judge’s impartialityinto question. See 28 U.S.C. § 455(a); Inre School Asbestos Litig., 977 F.2d 764, 783(3d Cir.1992). Section 455 does not pre-scribe the scope of disqualification. Rath-er, Congress ‘‘delegated to the judiciarythe task of fashioning the remedies thatwill best serve the purpose’’ of the disqual-ification statute. Liljeberg, 486 U.S. at862, 108 S.Ct. 2194.

[77] At a minimum, § 455(a) requiresprospective disqualification of the offend-ing judge, that is, disqualification from thejudge’s hearing any further proceedings inthe case. See United States v. MicrosoftCorp., 56 F.3d 1448, 1463–65 (D.C.Cir.1995) (per curiam) (‘‘Microsoft I’’). Micro-soft urges retroactive disqualification ofthe District Judge, which would entail dis-qualification antedated to an earlier part ofthe proceedings and vacatur of all subse-quent acts. Cf. In re School Asbestos Li-tig., 977 F.2d at 786 (discussing remedyoptions).

[78] ‘‘There need not be a draconianremedy for every violation of § 455(a).’’Liljeberg, 486 U.S. at 862, 108 S.Ct. 2194.Liljeberg held that a district judge couldbe disqualified under § 455(a) after enter-ing final judgment in a case, even thoughthe judge was not (but should have been)aware of the grounds for disqualificationbefore final judgment. The Court identifiedthree factors relevant to the questionwhether vacatur is appropriate: ‘‘in deter-mining whether a judgment should be va-cated for a violation of § 455(a), it is ap-propriate to consider the risk of injustice

to the parties in the particular case, therisk that the denial of relief will produceinjustice in other cases, and the risk ofundermining the public’s confidence in thejudicial process.’’ Id. at 864, 108 S.Ct.2194. Although the Court was discussing§ 455(a) in a slightly different context (thejudgment there had become final after ap-peal and the movant sought to have itvacated under Rule 60(b)), we believe thetest it propounded applies as well to casessuch as this in which the full extent of thedisqualifying circumstances came to lightonly while the appeal was pending. See Inre School Asbestos Litig., 977 F.2d at 785.

Our application of Liljeberg leads us toconclude that the appropriate remedy forthe violations of § 455(a) is disqualificationof the District Judge retroactive only tothe date he entered the order breaking upMicrosoft. We therefore will vacate thatorder in its entirety and remand this caseto a different District Judge, but will notset aside the existing Findings of Fact orConclusions of Law (except insofar as spe-cific findings are clearly erroneous or legalconclusions are incorrect).

This partially retroactive disqualificationminimizes the risk of injustice to the par-ties and the damage to public confidence inthe judicial process. Although the viola-tions of the Code of Conduct and § 455(a)were serious, full retroactive disqualifica-tion is unnecessary. It would unduly pe-nalize plaintiffs, who were innocent andunaware of the misconduct, and wouldhave only slight marginal deterrent effect.

Most important, full retroactive disquali-fication is unnecessary to protect Micro-soft’s right to an impartial adjudication.The District Judge’s conduct destroyed theappearance of impartiality. Microsoft nei-ther alleged nor demonstrated that it roseto the level of actual bias or prejudice.There is no reason to presume that every-thing the District Judge did is suspect.


See In re Allied–Signal Inc., 891 F.2d 974,975–76 (1st Cir.1989); cf. Liberty Lobby,Inc. v. Dow Jones & Co., 838 F.2d 1287,1301–02 (D.C.Cir.1988). Although Micro-soft challenged very few of the findings asclearly erroneous, we have carefully re-viewed the entire record and discern nobasis to suppose that actual bias infectedhis factual findings.

The most serious judicial misconduct oc-curred near or during the remedial stage.It is therefore commensurate that ourremedy focus on that stage of the case.The District Judge’s impatience with whathe viewed as intransigence on the part ofthe company; his refusal to allow an evi-dentiary hearing; his analogizing Micro-soft to Japan at the end of World War II;his story about the mule—all of these out-of-court remarks and others, plus theJudge’s evident efforts to please the press,would give a reasonable, informed observ-er cause to question his impartiality inordering the company split in two.

To repeat, we disqualify the DistrictJudge retroactive only to the imposition ofthe remedy, and thus vacate the remedyorder for the reasons given in Section Vand because of the appearance of partialitycreated by the District Judge’s miscon-duct.

2. Review of Findings of Fact andConclusions of Law

[79] Given the limited scope of our dis-qualification of the District Judge, we havelet stand for review his Findings of Factand Conclusions of Law. The severity ofthe District Judge’s misconduct and theappearance of partiality it created have ledus to consider whether we can and shouldsubject his factfindings to greater scrutiny.For a number of reasons we have rejectedany such approach.

The Federal Rules require that districtcourt findings of fact not be set aside

unless they are clearly erroneous. See FED.

R.CIV.P. 52(a). Ordinarily, there is no ba-sis for doubting that the District Court’sfactual findings are entitled to the substan-tial deference the clearly erroneous stan-dard entails. But of course this is noordinary case. Deference to a districtcourt’s factfindings presumes impartialityon the lower court’s part. When impar-tiality is called into question, how muchdeference is due?

[80] The question implies that there issome middle ground, but we believe thereis none. As the rules are written, districtcourt factfindings receive either full defer-ence under the clearly erroneous standardor they must be vacated. There is no denovo appellate review of factfindings andno intermediate level between de novo andclear error, not even for findings the courtof appeals may consider sub-par. SeeAmadeo v. Zant, 486 U.S. 214, 228, 108S.Ct. 1771, 100 L.Ed.2d 249 (1988) (‘‘TheDistrict Court’s lack of precision, however,is no excuse for the Court of Appeals toignore the dictates of Rule 52(a) and en-gage in impermissible appellate factfind-ing.’’); Anderson v. City of Bessemer City,470 U.S. 564, 571–75, 105 S.Ct. 1504, 84L.Ed.2d 518 (1985) (criticizing districtcourt practice of adopting a party’s pro-posed factfindings but overturning court ofappeals’ application of ‘‘close scrutiny’’ tosuch findings).

Rule 52(a) mandates clearly erroneousreview of all district court factfindings:‘‘Findings of fact, whether based on oral ordocumentary evidence, shall not be setaside unless clearly erroneous, and dueregard shall be given to the opportunity ofthe trial court to judge of the credibility ofthe witnesses.’’ FED.R.CIV.P. 52(a). Therule ‘‘does not make exceptions or purportto exclude certain categories of factualfindings from the obligation of a court of


appeals to accept a district court’s findingsunless clearly erroneous.’’ Pullman-Stan-dard v. Swint, 456 U.S. 273, 287, 102 S.Ct.1781, 72 L.Ed.2d 66 (1982); see alsoAnderson, 470 U.S. at 574–75, 105 S.Ct.1504; Inwood Labs., Inc. v. Ives Labs.,Inc., 456 U.S. 844, 855–58, 102 S.Ct. 2182,72 L.Ed.2d 606 (1982). The SupremeCourt has emphasized on multiple occa-sions that ‘‘[i]n applying the clearly errone-ous standard to the findings of a districtcourt sitting without a jury, appellatecourts must constantly have in mind thattheir function is not to decide factual is-sues de novo.’’ Zenith Radio Corp. v.Hazeltine Research, Inc., 395 U.S. 100,123, 89 S.Ct. 1562, 23 L.Ed.2d 129 (1969);Anderson, 470 U.S. at 573, 105 S.Ct. 1504(quoting Zenith).

[81] The mandatory nature of Rule52(a) does not compel us to accept factfind-ings that result from the District Court’smisapplication of governing law or thatotherwise do not permit meaningful appel-late review. See Pullman–Standard, 456U.S. at 292, 102 S.Ct. 1781; Inwood Labs.,456 U.S. at 855 n. 15, 102 S.Ct. 2182. Normust we accept findings that are utterlydeficient in other ways. In such a case, wevacate and remand for further factfinding.See 9 MOORE’S FEDERAL PRACTICE § 52.12[1](Matthew Bender 3d ed.2000); 9ACHARLES A. WRIGHT & ARTHUR R. MILLER,

FEDERAL PRACTICE AND PROCEDURE § 2577,at 514–22 (2d ed.1995); cf. Icicle Seafoods,Inc. v. Worthington, 475 U.S. 709, 714, 106S.Ct. 1527, 89 L.Ed.2d 739 (1986); Pull-man-Standard, 456 U.S. at 291–92, 102S.Ct. 1781.

[82] When there is fair room for argu-ment that the District Court’s factfindingsshould be vacated in toto, the court ofappeals should be especially careful in de-termining that the findings are worthy ofthe deference Rule 52(a) prescribes. See,e.g., Thermo Electron Corp. v. Schiavone

Constr. Co., 915 F.2d 770, 773 (1st Cir.1990); cf. Bose Corp. v. Consumers Unionof United States, Inc., 466 U.S. 485, 499,104 S.Ct. 1949, 80 L.Ed.2d 502 (1984).Thus, although Microsoft alleged only ap-pearance of bias, not actual bias, we havereviewed the record with painstaking careand have discerned no evidence of actualbias. See S. Pac. Communications Co. v.AT & T, 740 F.2d 980, 984 (D.C.Cir.1984);Cooley, 1 F.3d at 996 (disqualifying districtjudge for appearance of partiality but not-ing that ‘‘the record of the proceedingsbelow TTT discloses no bias’’).

In light of this conclusion, the DistrictJudge’s factual findings both warrant def-erence under the clear error standard ofreview and, though exceedingly sparing incitations to the record, permit meaningfulappellate review. In reaching these con-clusions, we have not ignored the DistrictJudge’s reported intention to craft his fact-findings and Conclusions of Law to mini-mize the breadth of our review. TheJudge reportedly told Ken Auletta that‘‘[w]hat I want to do is confront the Courtof Appeals with an established factual rec-ord which is a fait accompli.’’ AULETTA,

WORLD WAR 3.0, at 230. He explained:‘‘part of the inspiration for doing that isthat I take mild offense at their reversal ofmy preliminary injunction in the consent-decree case, where they went ahead andmade up about ninety percent of the factson their own.’’ Id. Whether the DistrictJudge takes offense, mild or severe, isbeside the point. Appellate decisions com-mand compliance, not agreement. We donot view the District Judge’s remarks asanything other than his expression of dis-agreement with this court’s decision, andhis desire to provide extensive factual find-ings in this case, which he did.

VII. CONCLUSION

The judgment of the District Court isaffirmed in part, reversed in part, and

119HONEYWELL INTERN., INC. v. N.L.R.B.Cite as 253 F.3d 119 (D.C. Cir. 2001)

remanded in part. We vacate in full theFinal Judgment embodying the remedialorder, and remand the case to the DistrictCourt for reassignment to a different trialjudge for further proceedings consistentwith this opinion.

,

HONEYWELL INTERNATIONAL,INC.,Petitioner,

v.

NATIONAL LABOR RELATIONSBOARD, Respondent.

International Union, United Automo-bile, Aerospace & Agricultural Imple-ment Workers of America, UAW, Lo-cal 376, Intervenor.

Nos. 00–1170 & 00–1274.

United States Court of Appeals,District of Columbia Circuit.

Argued May 8, 2001.

Decided June 29, 2001.

A petition was filed for review and across-application was filed for enforcementof an order of the National Labor Rela-tions Board (NLRB) finding that an em-ployer committed an unfair labor practice.The Court of Appeals, Randolph, CircuitJudge, held that the NLRB exceeded itsauthority by injecting itself into a contractdispute when the employer had a soundarguable basis for its interpretation.

Petition for review granted and deci-sion set aside.

1. Labor Relations O510

Where employer and union enteredinto a ‘‘Competitiveness Agreement’’ whichwas one of three agreements among themand which provided that employer couldterminate it if specified financial arrange-ments for the plant in question were notmade in the federal budget, the NationalLabor Relations Board (NLRB) exceededits authority by injecting itself into a con-tract dispute when it considered whetheremployer had made sufficient efforts toobtain funding and whether further effortswere futile, on theory the employer hadrepudiated the agreement in its entirety,and when the employer had a sound argua-ble basis for its interpretation. NationalLabor Relations Act, § 8(a)(5), as amend-ed, 29 U.S.C.A. § 158(a)(5); Labor Man-agement Relations Act, 1947, § 301, 29U.S.C.A. § 185.

2. Labor Relations O510The National Labor Relations Board

(NLRB) is empowered to interpret collec-tive bargaining agreements when doing sois necessary to determine whether an un-fair labor practice has occurred, but thefederal courts, not the Board, are legisla-tively empowered to be the primary inter-preters of contracts. Labor ManagementRelations Act, 1947, § 301, 29 U.S.C.A.§ 185.

3. Labor Relations O389An employer could not be deemed to

have refused to bargain collectively over aparticular subject when it had bargainedover that subject matter and memorial-ized the results in a contract, though itlater terminated the contract pursuant toits interpretation of its terms; once theNational Labor Relations Board (NLRB)determined that the contract covered amandatory subject of bargaining, its inter-pretive task was at an end, and if theparties wished to enforce their contract,

Documents

34 253 FEDERAL REPORTER, 3d SERIES - Berkeley Law · 34 253 FEDERAL REPORTER, 3d SERIES UNITED STATES of America, Appellee, v. MICROSOFT CORPORATION, Appellant. Nos. 00–5212 and