PREGERSON, Circuit Judge:
Appellant Transamerica Computer Company (Transamerica), a wholly owned subsidiary of Transamerica Corporation, alleges that Appellee International Business Machines (IBM) violated Section 2 of the Sherman Act, 15 U.S.C. § 2, when it took various actions to combat emerging competition in the “plug-compatible” peripherals market. The district court held that IBM’s actions did not violate the antitrust laws.
On appeal, Transamerica challenges the district court’s ruling that IBM’s acts did not “unreasonably restrict” competition and, in particular, challenges the court’s test for predatory pricing. We affirm the district court’s decision but modify its test for predatory pricing.
BACKGROUND
At the heart of a computer system is the central processing unit (CPU), which houses arithmetical and logical electronic circuits. Attached to the CPU are devices called “peripherals,” which perform input, output, storage, and control functions. IBM, long the dominant force in the computer industry, was, and remains, the major supplier of both CPUs and peripherals.
In 1967, a number of companies began offering plug-compatible peripherals — devices which could be attached to IBM’s CPUs. These plug-compatible manufacturers (PCMs) enjoyed immediate market success because they offered, at substantial discounts, plug-compatible peripherals that performed as well as or better than IBM’s peripherals.
Transamerica was formed to supply needed capital to PCMs. In financing transactions central to this case, Transamerica purchased millions of dollars of peripherals from two PCMs, Marshall Industries and Telex Corporation, which had previously leased these items to end users. Under this arrangement, these PCMs raised substantial capital, and Transamerica, in addition to acquiring the equipment and underlying leases, received substantial tax advantages.
IBM responded to vigorous competition from the PCMs by engaging in a number of programs which Transamerica characterizes as violations of the Sherman Act. These challenged programs included:
(1) Leasing Program — Prior to May 1971, IBM customers could either buy peripheral equipment or lease it on a month-to-month basis. In May 1971, IBM announced an additional method of leasing peripheral equipment — a Fixed-Term Lease Plan under which customers could lease peripheral equipment for one year at an eight percent discount below the month-to-month rate, or for two years at a sixteen percent discount.
(2) Design Changes — In the early 1970s, IBM redesigned the interface between the CPU and the peripherals of three tape drive systems so that PCM’s peripherals would no longer be [1381]*1381compatible with IBM’s CPUs. IBM also removed an optional selector channel from two CPU models, the System 370 Models 115 and 125, so that PCM’s peripherals could no longer be used with those models.
(3) Pricing Behavior — Also in the early 1970s, IBM introduced several “new” products — basically repackaged versions of prior peripherals — at lower prices.
Whether one characterizes IBM’s actions as “meeting” competition or “precluding” competition, there is no doubt that IBM’s strategy worked. Transamerica, along with fifteen out of seventeen companies involved with plug-compatible peripherals, left the market after suffering huge losses.
Because of IBM’s actions in the peripherals market, Transamerica sued IBM for violations of the antitrust laws. Several other companies involved with plug-compatible peripherals also brought suits against IBM. Although initially consolidated, the actions were tried separately. Two of those actions, involving issues and facts similar to those presented here, have already been before this court. In both instances we upheld directed verdicts for IBM. California Computer Products, Inc. v. IBM, 613 F.2d 727 (9th Cir.1979) (CalComp); Memorex Corp. v. IBM, 636 F.2d 1188 (9th Cir.1980), cert. denied, 452 U.S. 972, 101 S.Ct. 3126, 69 L.Ed.2d 983 (1981) (Memorex).
The instant case went to trial before a jury in December 1978. The trial consumed 120 days. After deliberating for ten days, the jury deadlocked and was discharged. Under a pretrial stipulation, the district judge then became the trier of fact. He ruled for IBM on all major issues. He held that (1) IBM was not a monopolist in either the general computer systems market or the peripherals market; (2) assuming, arguendo, that IBM possessed monopoly power, its leasing program, its design changes— with the exception of the design of the System 370 Models 115 and 125 — and its pricing behavior did not unreasonably restrain competition; (3) IBM had not attempted to monopolize the general computer systems market or the peripherals market; and (4) assuming that Transamerica established antitrust liability, it had not proved that it suffered antitrust damages. The district court also found that IBM’s redesign of the System 370 Models 115 and 125 would have unreasonably restricted competition had IBM been a monopolist, but since IBM was not a monopolist and since Transamerica suffered no damages resulting from that design change, IBM’s redesign of the two CPUs did not render IBM liable for antitrust damages. In re IBM Peripheral EDP Devices Antitrust Litigation, Transamerica Computer Co. Inc. v. IBM, 481 F.Supp. 965 (N.D.Cal.1979) (Transamerica Computer).
STANDARD OF REVIEW
Transamerica challenges several of the district court’s findings of fact.1 The district court’s findings may not be reversed unless clearly erroneous. Fed.R.Civ.P. 52(a). Under this standard
[a] finding is “clearly erroneous” when although there is evidence to support it, the reviewing court on the entire evidence is left with the definite and firm conviction that a mistake has been committed.
United States v. United States Gypsum Co., 333 U.S. 364, 395, 68 S.Ct. 525, 541, 92 L.Ed. 746 (1948).
[1382]*1382MONOPOLIZATION AND ATTEMPT TO MONOPOLIZE
Transamerica charges that IBM either monopolized or attempted to monopolize certain segments of the computer market. These separate offenses are governed by different tests. To establish monopolization, a plaintiff must prove:
(1) possession of monopoly power in the relevant market;
(2) willful acquisition or maintenance of that power; and
(3) causal “antitrust” injury.
CalComp, 613 F.2d at 735.
To establish that a defendant attempted to monopolize, a plaintiff must prove:
(1) specific intent to control prices or destroy competition with respect to a part of commerce;
(2) predatory or anticompetitive conduct directed to accomplishing the unlawful purpose;
(3) a dangerous probability of success; and
(4) causal “antitrust” injury.
Id. at 736.
There is an important relationship between the second elements of these two offenses. Conduct that does not constitute “willful acquisition or maintenance” of monopoly power (thus precluding establishment of the offense of monopolization) cannot constitute the “predatory or anticompetitive conduct” required to establish the offense of attempt to monopolize. See Cal-Comp, 613 F.2d at 738, Quoting 3 P. Areeda & D. Turner, Antitrust Law If 828, at 321 (1978) (“conduct lawful for a monopolist must, a fortiori, be excluded as a basis for the attempt offense.”). We will analyze IBM’s conduct with this principle in mind. We will assume that IBM possessed monopoly power. If IBM’s conduct proves lawful despite that assumption, then, a fortiori, IBM’s conduct could not constitute an attempt to monopolize, thereby eliminating the need to consider this offense.
LEASING PRACTICES
Transamerica alleges that IBM’s Fixed Term Lease Plan (FTP) unreasonably restrained competition. Plaintiffs in both CalComp and Memorex challenged the legality of the FTP. Both times, this court upheld directed verdicts in favor of IBM on that issue. CalComp, 613 F.2d at 741-42; Memorex, 636 F.2d at 1188. The district court here also concluded that the FTP was legal. Transamerica Computer, 481 F.Supp. at 1001-02.
Transamerica asserts that this court’s previous decisions on the FTP are not controlling because they were based on findings of fact proved erroneous in the instant trial. The district court’s findings of fact, however, are indistinguishable from the findings in CalComp and Memorex. Transamerica has not demonstrated that the district court’s findings in the instant case are clearly erroneous. Therefore, the holdings of CalComp and Memorex control our disposition of this issue, and the district court’s holding must be affirmed. See also Greyhound Computer Corp. v. IBM, 559 F.2d 488, 498-99 (9th Cir.1977), cert. denied, 434 U.S. 1040, 98 S.Ct. 782, 54 L.Ed.2d 790 (1978).
DESIGN CHANGES
IBM’s design changes challenged in this case — redesign of the interface between the CPU and certain peripherals— are of the same type as those previously contested in CalComp and Memorex. Those cases upheld the legality of IBM’s design changes. The district court here made a similar ruling, finding that the contested changes were improvements in the products, were not unreasonably restrictive of competition, and hence did not violate the Sherman Act.2 Transamerica Computer, [1383]*1383481 F.Supp. at 1003-05. Transamerica has not demonstrated how the design changes in this case can be distinguished from the changes in the two other cases or how the district court’s findings are clearly erroneous. Again, on the authority of CalComp and Memorex, we affirm the district court’s ruling that the interface changes were not unreasonable. See also Telex Corp. v. IBM, 510 F.2d 894, 902, 906 (10th Cir.1975), rev’g 367 F.Supp. 258 (N.D.Okla.1973), cert. dismissed, 423 U.S. 802, 96 S.Ct. 8, 46 L.Ed.2d 244 (1975).
Transamerica also charges that IBM’s redesign of the System 370 Model 115 and 125 CPUs unreasonably restricted competition. The 115 and 125 models were the smallest of IBM’s System 370 CPUs. The models included a channel for attaching slower speed devices. The district court found that IBM redesigned the models to operate just short of the speed that would have enabled peripherals manufactured by PCMs to attach, and thus that the change unreasonably restricted competition. Transamerica Computer, 481 F.Supp. at 1006-08.
The district court, however, refused to award Transamerica damages for two reasons. First, it found that IBM did not possess monopoly power in the requisite market. Second, it found that Transamerica did not suffer any injury as a result of IBM’s conduct because
[t]he market for the tapes excluded by this conduct was insignificant. Only tape drives with data rates between 30 and 50KB were affected. Those were older, low performance technology devices that would only have been competitive at prices far below those contemplated in Transamerica’s ‘damage claim. Transamerica probably owned some of them, but its President didn’t know how many, and any Transamerica owned were purchased in the second Telex tape contract. Because that contract was not an arms-length transaction, this Court would be reluctant to award Transamerica any damages on the equipment purchased thereunder. Also, there is no evidence that Transamerica in the past supplied peripherals for the 115/125 migrator systems, or that it intended to, or took any steps to supply peripherals for the 115 and 125 systems.
Transamerica Computer, 481 F.Supp. at 1008 n. 109.
We need not consider the first of these findings — that IBM lacked monopoly power — because we hold that the district court’s finding that Transamerica suffered no damages attributable to the redesign of the Models 115 and 125 is not clearly erroneous. Without proving antitrust injury, Transamerica cannot recover for the antitrust violation if, in fact, any violation occurred. CalComp, 613 F.2d at 732. We affirm the district court’s holdings on IBM’s design changes.
PREDATORY PRICING
A. The District Court’s Findings
In response to the challenge from the PCMs, IBM introduced several “new” products which actually were repackaged versions of existing products. The “new” products were priced below the older versions. Transamerica asserts that these lower prices were predatory.
The district court carefully examined these price cuts. Transamerica Computer, 481 F.Supp. at 996-1002. It concluded that IBM expected the new products to “return substantial profits,” id. at 997; that, in fact, the products were profitable; and that the prices at issue exceeded the average total cost of producing the products. Id. at 1002. After an extensive analysis, the court concluded that IBM’s pricing policy was legal.
The district court’s finding that the challenged prices exceeded IBM’s average cost is not clearly erroneous.3 The district [1384]*1384court, however, applied an incorrect test in deciding whether such prices are predatory. We set out the correct test below, apply that test to this case, and conclude that IBM’s pricing policy was legal.
B. The Economic and Legal Background
We differ with the district court on the proper analysis of situations where a defendant’s prices are alleged to be predatory even though they exceed the defendant’s average total cost. To analyze such situations, a preliminary discussion of economic terminology and legal precedents is helpful.
Predatory pricing occurs when a company that controls a substantial market share lowers its prices to drive out competition so that it can charge monopoly prices, and reap monopoly profits, at a later time. William Inglis & Sons Baking Co. v. ITT Continental Baking Co., Inc., 668 F.2d 1014, 1031-32 (9th Cir.1981), cert. denied, - U.S. -, 103 S.Ct. 58, 74 L.Ed.2d 61 (1982); Janich Bros., Inc. v. American Distilling Co., 570 F.2d 848, 856 (9th Cir.1977), cert. denied, 439 U.S. 829, 99 S.Ct. 103, 58 L.Ed.2d 122 (1978); L. Sullivan, Handbook of the Law of Antitrust 109 (1977). It can be difficult, however, to distinguish predatory price cuts intended to eliminate competition from legitimate price cuts designed to meet or beat competition. An influential attempt to clarify the distinction was made by Professors Phillip Areeda and Donald Turner, who proposed a “cost-based” test for predation.4 Under such a test, the relation between the cost of producing a product and the price charged for „ it is the criterion for determining whether the price is predatory.
Economists, however, measure a firm’s costs in a number of ways, and these measures are relevant in understanding the Areeda-Turner proposal and this court’s reaction to it. Costs are divided into fixed costs (those that do not vary with changes in output) and variable costs (which do so vary). Total cost is the sum of fixed and variable costs. Marginal cost is the increment to total cost that results from producing an additional unit of output. Average cost, or average total cost, is obtained by dividing total cost by output. Likewise, average variable cost is the sum of all variable costs divided by output. Average cost [1385]*1385is thus higher than average variable cost for all output levels.5
Areeda and Turner suggest that prices be considered per se lawful (i.e., non-predatory) if they exceed the defendant’s marginal cost or average variable cost,6 and that prices be considered per se illegal (predatory) if they are below marginal or average variable cost.7 The rationale for this cost-based per se test is the belief that a company that makes a profit, however small, on each additional product does so because it is efficient, and should not face antitrust challenges, whereas a company that loses money on the sale of an additional product is doing so presumably for anti-competitive reasons.
The Areeda-Turner test has provoked much judicial and academic comment.8 This court has been influenced by the Areeda-Turner test without unqualifiedly embracing it. In a series of opinions during the three years preceding the district court’s decision reviewed here,9 we approved the use of marginal or average variable cost in establishing predation without making that mode of proof exclusive, as Areeda and Turner advocate. Indeed, we adopted neither the Areeda-Turner test’s conclusive presumption that prices above marginal or average variable cost are legal, nor its conclusive presumption that prices below that cut-off point are predatory. It is true that in Hanson v. Shell Oil Co., 541 [1386]*1386F.2d 1352, 1359 n. 6 (9th Cir.1976), cert. denied, 429 U.S. 1074, 97 S.Ct. 813, 50 L.Ed.2d 792 (1977), we spoke of “proof of pricing below marginal or average variable cost” as a “prerequisite to a prima facie showing of an attempt to monopolize.” Nevertheless, we suggested that prices above marginal or average variable cost might, in appropriate circumstances, be found predatory, id. at 1358 n. 5, and we reiterated that point in CalComp, 613 F.2d at 743. We also explicitly denied that prices below marginal or average variable cost were per se unlawful. Hanson, 541 F.2d at 1359 n. 6. In short, the closest we were willing to move to the Areeda-Turner approach was to acknowledge that a “price set at or above marginal cost should not ordinarily form the basis for an antitrust violation.” Janich Bros., Inc. v. American Distilling Co., 570 F.2d at 857 (footnote omitted) (emphasis added).
Any doubt that this circuit rejects the per se aspects of the Areeda-Turner test was dispelled in William Inglis & Sons Baking Co. v. ITT Continental Baking Co., Inc., 668 F.2d 1014 (9th Cir.1981), cert. denied, -U.S.-, 103 S.Ct. 58, 74 L.Ed.2d 61 (1982), decided after the district court’s decision in the instant case. In Inglis we explained that prices are predatory when their justification rests, “not on their effectiveness in minimizing losses, but on their tendency to eliminate rivals and create a market structure enabling the seller to recoup his losses.” Id. at 1035.10 We emphasized that this standard, “and not rigid adherence to a particular cost-based rule ... must govern our analysis of alleged predatory pricing.” Id. Yet Inglis did not, as Transamerica contends here, repudiate all cost-based tests. Rather, it laid down a cost-based test for allocating the burden of proof on the predation issue in place of one designed (like the Areeda-Turner test) to resolve that issue conclusively:
[T]o establish predatory pricing a plaintiff must prove that the anticipated benefits of defendant’s price depended on its tendency to discipline or eliminate competition and thereby enhance the firm’s long-term ability to reap the benefits of monopoly power. If the defendant’s prices were below average total cost but above average variable cost, the plaintiff bears the burden of showing defendant’s pricing was predatory. If, however, the plaintiff proves that the defendant’s prices were below average variable cost, the plaintiff has established a prima facie case of predatory pricing and the burden shifts to the defendant to prove that the prices were justified without regard to any anticipated destructive effect- they might have on competitors.
Id. at 1035-36.
C. The Treatment of Prices Above Average Total Cost
The Inglis test, quoted above, addresses only two categories of prices — those below average variable cost (which, under burden shifting, the defendant must prove are non-predatory)11 and those above average variable cost but below average total cost (which the plaintiff has the burden of proving are predatory). The Inglis test says nothing about how to evaluate prices for antitrust purposes that exceed average total cost. Indeed, Inglis explicitly left this question open. 668 F.2d at 1035 n. 30.
In the instant case, the district court, which did not have the benefit of Inglis and its clarification of our previous discussions of the Areeda-Turner test, held that prices above average total cost “should be conclusively presumed legal.” Transamerica Computer, 481 F.Supp. at 991.12 We disagree for several reasons.
[1387]*1387First, this court has already recognized that prices exceeding average total cost might nevertheless be predatory in some circumstances. The specific example we discussed was “limit pricing,” in which a monopolist sets prices above average total cost but below the short-term profit-maximizing level so as to discourage new entrants and thereby maximize profits over the long run. See 3 P. Areeda & D. Turner, supra U 714b. We explained that “limit pricing by a monopolist might, on a record which presented the issue, be held an impermissible predatory practice.” GalComp, 613 F.2d at 743. A similar pricing strategy would be for a monopolist to make temporary reductions to a level above average total cost but below the profit-maximizing price whenever a new entrant appears ready to enter the market. One or two such reductions could discourage potential entrants in a market that requires sizable initial investments, leaving the monopolist free to raise his prices to monopoly levels. See 3 P. Areeda & D. Turner, supra 714c. Such a pricing strategy, like limit pricing, could well be found predatory.13
Second, the district court’s per se test rests on the notion that price reductions to average total cost result from efficient production and harm only less efficient competitors. Transamerica Computer, 481 F.Supp. at 991. But companies may lower their prices for temporary strategic reasons as well. One critic of the AreedaTurner test points out that a monopolist can employ price strategies that jeopardize consumers’ long-run welfare without lowering prices below average total cost and concludes that “it is unrealistic and even analytically wrong to apply a simple short-run price-cost rule for determining whether exclusionary pricing by a monopolist is socially undesirable and therefore predatory.” Scherer, Predatory Pricing and the Sherman Act: A Comment, 89 Harv.L.Rev. 869, 890 (1976). It may be difficult in many or most instances to assess the long-run consequences of challenged pricing policies.14 But where those difficulties can be overcome, the law should not prevent plaintiffs from proving antitrust violations.
Third, the uncertainty and imprecision inherent in determining “costs” counsel against basing conclusive presumptions on the relation between prices and costs. Assessing those relations for the products of a multi-product firm requires allocating known and estimated costs and revenues among various products. While accounting problems do not warrant ignoring cost figures completely, they do make it unwise to rely exclusively on such figures.
Finally, we should hesitate to create a “free zone” in which monopolists can exploit their power without fear of scrutiny by the law. A rule based exclusively on cost forecloses consideration of other important factors, such as intent, market power, market structure, and long-run behavior in evaluating the predatory impact of a pricing decision.15
[1388]*1388For these reasons, we disagree with the district court’s conclusion that prices above average total cost should be legal per se. Rather, we believe that Inglis adopted the proper approach to the use of cost figures in determining whether prices are predatory: cost categories should be used to allocate the burden of proof on the issue of predation. By this approach, we give due weight to the economic considerations which suggest that prices are presumptively lawful if they exceed marginal or average variable cost and presumptively predatory if they do not. And, of course, this approach does not preclude a litigant from introducing evidence sufficient to overcome these presumptions. Inglis followed this approach in evaluating prices below average variable cost and prices between average variable cost and average total cost. The logic of the Inglis approach applies with equal force in evaluating prices above average total cost.
The test for determining the antitrust legality of prices that exceed average total cost should be consistent with the Inglis approach and with our view that cost-price relations should not be the exclusive method of proving predation. In addition, the test should be consistent with the economic analysis of Areeda and Turner. Their analysis indicates that prices above average total cost will rarely be predatory.16 Therefore, it is appropriate to impose on the plaintiff a greater burden of proving that prices above average total cost are predatory than the burden imposed by Inglis to prove that prices between average variable and average total cost are predatory. We therefore hold that if the challenged prices exceed average total cost, the plaintiff must prove by clear and convincing evidence— i.e., that it is highly probably true — that the defendant’s pricing policy was predatory.
D. The Proper Test Applied to This Case.
While we modify the test applied by the district court, we affirm the court’s decision that IBM’s pricing policy did not violate the antitrust laws.17
The district court found that IBM’s prices were above its average total cost. Transamerica Computer, 481 F.Supp. at 1002. This finding was not clearly erroneous. Thus, to prevail on its claim of predatory pricing, Transamerica must prove by clear and convincing evidence that IBM’s pricing policy unreasonably restricted competition. The exhaustive trial record, containing a plethora of evidence on pricing behavior, market structure, production costs, marketing strategies, and other related information, fails to provide any clear and convincing evidence of predatory pricing.18 Transamerica did not, for example, introduce evidence that IBM’s prices rose once competition had left the market. The only price increases cited by Transamerica were modest increases in the mid-1970s. During that [1389]*1389inflationary period, however, IBM increased prices on all its products, not just those involved in this ease. Transamerica did not prove that IBM engaged in limit pricing.19 Transamerica simply introduced evidence of an initial price cut, hardly an unusual act in the computer industry or unusual in the face of competition. Transameriea’s evidence of predatory pricing falls far short of the type of clear and convincing evidence that would permit the trier of fact to find predatory pricing when prices are above average' total cost.
CONCLUSION
Thus, even assuming that IBM possessed monopoly power in the relevant market, its lease plan, design changes, and pricing policy did not constitute unreasonable restrictions on competition. On this record, IBM was entitled to judgment as a matter of law. Therefore, we need not consider whether the district court erred in finding that IBM did not possess monopoly power, Transamerica Computer, 481 F.Supp. at 974-87, or that Transamerica had failed to prove damages. Id. at 1010-21.
The judgment is AFFIRMED.