Opinion for the court filed by Circuit Judge BRYSON. Dissenting opinion filed by Circuit Judge PAULINE NEWMAN.
BRYSON, Circuit Judge.I
The polymerase chain reaction (PCR) allows scientists, beginning with a small amount of deoxyribonucleic acid (DNA), to generate many copies of that DNA in a *1358short period of time. The ability of PCR to synthesize DNA rapidly has led to significant advances in molecular biology and has been particularly useful in pathology and in the identification of trace materials such as blood and hair.
In the first stage of PCR, a segment of DNA is separated at high temperature into its two component strands. Then, at a lower temperature, small pieces of synthetic DNA called “primers” are annealed to specific locations on the separated strands. Enzymes known as DNA polymerases then “extend” the primers by attaching a complementary nucleotide to each nucleotide in the template strand. In this manner, the polymerase creates two identical double-stranded DNA helices from the two separated single strands of the original helix. The process of strand separation, primer annealment, and extension is then performed repeatedly, resulting in the production of a large number of identical DNA strands.
When PCR was first developed, the high temperatures associated with strand separation destroyed the polymerase that was used to drive the reaction. Accordingly, new polymerase had to be added at the beginning of each cycle of the reaction, which was cumbersome. It was subsequently discovered that the DNA polymerase of the Thermus aquaticus, or “Taq,” bacterium, which is found in geysers and hot springs, was stable and active at high temperatures and therefore could withstand the rigors of PCR. Thus, Taq needed to be added to the reaction mixture only once, which resulted in making the PCR process much faster and more efficient.
On June 17, 1987, Cetus Corporation, the predecessor of appellants Hoffmann-La Roche, Inc., and Roche Molecular Systems, Inc. (collectively, “Roche”), filed U.S. Patent Application No. 07/063,509 (“the '509 application”), which was directed to a purified thermostable enzyme. The '509 application named Dr. David Gelfand, the Cetus scientist most knowledgeable about the Taq enzyme, and Ms. Susanne Stoffel, Dr. Gelfand’s technician, as inventors. The broadest originally filed claim was not limited to the Taq enzyme.
In an office action dated November 1, 1988, the examiner rejected all the submitted claims on a variety of grounds. The rejections included an anticipation/obviousness rejection based on journal articles by Chien, et al., and Kaledin, et al., both of which disclosed a DNA polymerase derived from the Taq bacterium. The examiner noted that the applicants included molecular weight limitations in dependent claims and that those limitations differed from the estimates of Taq’s molecular weight reported by Chien and Kaledin. The examiner suggested that “some proteins behave anomalously when subjected to SDS page,” the technique used by the applicants to estimate molecular weight. For that reason, the examiner concluded that “[i]t is not clear whether or not the molecular weight an[d] pH range of activity claimed by applicants for the instant enzyme is a result of experimental parameters or an enzyme activity different than the [enzyme] previously described in the literature.”
On March 17, 1989, the inventors responded to the examiner’s rejection, canceling all pending claims and entering three new claims, the broadest of which provided as follows:
1. Purified thermostable Thermus aquaticus DNA polymerase that migrates on a denaturing polyacrylamide gel faster than phosphorylase B and more slowly than does bovine serum albumin and has an estimated molecular weight of 86,000-90,000 daltons when compared with a phosphorylase B stan*1359dard assigned a molecular weight of 92,-500 daltons.
U.S. Patent No. 4,889,818, col. 44,ll. 46-52.
In remarks accompanying the amendment, the applicants made a two-part argument for patentability. First, they asserted that the claimed enzyme was distinct from the prior art enzyme, citing differences in molecular weight, specific activity, and fidelity. Second, they contended that even if, contrary to their belief, the claimed and prior art enzymes were identical, the claimed enzyme would still be patentable because it was “far more pure” than the enzyme of the Chien and Kaledin preparations. To support that assertion, the inventors cited a portion of the application indicating that the claimed enzyme had a specific activity ten times that of the prior art enzyme.
The examiner allowed the amended claims without further comment. The '509 application therefore issued as U.S. Patent No. 4,889,818 (“the '818 patent”) on December 26,1989.
Cetus licensed the '818 patent to Prome-ga Corporation in June 1990. After Cetus sold the '818 patent to Roche, Promega allegedly breached the license agreement. Roche filed suit, alleging patent infringement and breach of contract. Promega counterclaimed, asserting inter alia that the '818 patent was unenforceable due to inequitable conduct, a claim that soon became the focus of the litigation.
In August 1996, the district court held on summary judgment that the inventors had made four material misrepresentations during the prosecution of the '818 patent. After a bench trial, the court held that the '818 patent was unenforceable based on eight separate misrepresentations and omissions, including three of the ones the court had addressed at the summary judgment stage. Roche appeals the court’s order holding the patent unenforceable.
II
A party seeking to have a patent declared unenforceable has a heavy burden to meet. Inequitable conduct requires misrepresentation or omission of a material fact, together with an intent to deceive the PTO. Both of those distinct elements must be shown by clear and convincing evidence. See Manville Sales Corp. v. Paramount Sys., Inc., 917 F.2d 544, 552, 16 USPQ2d 1587, 1593 (Fed.Cir.1990); Kingsdown Med. Consultants, Ltd. v. Hollister Inc., 863 F.2d 867, 872, 9 USPQ2d 1384, 1389 (Fed.Cir.1988). Once the requisite levels of materiality and intent are shown, the district court must determine whether the equities warrant a conclusion that the patentee has engaged in inequitable conduct. Molins PLC v. Textron, Inc., 48 F.3d 1172, 1178, 33 USPQ2d 1823, 1827 (Fed.Cir.1995).
While it is difficult to prove inequitable conduct, a district court’s ruling on inequitable conduct is reviewed deferentially. The court’s findings on materiality and intent are reviewed for clear error, and thus will not be overturned in the absence of a “definite and firm conviction” that a mistake has been made. Molins, 48 F.3d at 1178, 33 USPQ2d at 1827. The district court’s assessment of the equities is reviewed for an abuse of discretion. Id.
The district court in this case based its inequitable conduct ruling on its finding that the inventors made various material misrepresentations and related omissions, which can be grouped into three categories: (1) representations regarding the difference in molecular weight between the claimed and prior art Taq enzymes; (2) representations that the inventors had performed Example VI, one of the procedures described in the specification, and that they had achieved the described results; and (3) representations concerning the comparative fidelity and template depen*1360dence of the claimed enzyme and the prior art enzymes.
We analyze each of the three categories of asserted inequitable conduct below, upholding the district court’s findings as to two of the three and overturning the court’s findings as to the third. The dissent, which would overturn the district court’s findings with respect to all three categories, criticizes us for failing to apply the clear and convincing standard of proof in assessing the evidence of materiality and intent. While we recognize the standard of proof that applies to those factual issues, we also note that our task as a reviewing court is not to make those factual determinations in the first instance but to determine whether the district court’s findings on those issues are clearly erroneous or are infected with legal error. With that standard of review in mind, we turn to each of the grounds on which the district court based its ruling.
A. Molecular Weight
In their response to the examiner’s office action, the inventors argued that the claimed enzyme was different from the enzyme described in the cited references. Their argument focused principally on the disparity in the molecular weight values reported for the prior art enzymes and the claimed enzyme. The inventors noted that claim 1 of the '509 application recited an enzyme with a molecular weight of 86,000 to 90,000 daltons, while the Chien and Kaledin references described enzymes with molecular weights of 63,000-68,000 daltons and 60,000-62,000 daltons, respectively. The inventors proffered an explanation for the discrepancy: they believed that, at most, Chien and Kaledin had isolated a “crude preparation of degraded Taq polymerase.” The district court, however, concluded on summary judgment that the inventors had withheld information from the PTO that tended to show that the enzymes Chien and Kaledin had isolated were not simply degraded fragments of Taq polymerase and thus made it less clear that the Chien and Kaledin enzymes were distinguishable from the enzyme described in the application.
1. The Stoffel Experiment
The dispute over the disparity in reported molecular weights relates in part to an experiment that Ms. Stoffel performed, in which she applied a protein sample to a phosphocellulose column. She observed that a Taq fragment with a molecular weight of 62,000 daltons did not bind to the phosphocellulose column. According to the district court, the results of that experiment undercut the inventors’ statement in the March 17, 1989, office action response that the Chien article referred to a degraded form of Taq. In the district court’s view, the fact that the Taq fragment did not bind to Ms. Stoffel’s phosphocellulose column indicated that a similar fragment would have been “lost” in the experiment described by Chien, because that experiment also used a column of phosphocellu-lose. The district court therefore concluded that the Stoffel experiment suggested that the enzyme described in the prior art references was not a degraded form of Taq, as the inventors contended. For that reason, the court determined that the inventors should have replicated the prior art experiment before characterizing it as producing a Taq fragment with a lower molecular weight.
Roche argues that there were genuine issues of material fact regarding the materiality of the Stoffel experiment, and that the issue should not have been decided on summary judgment. We agree. Two key facts are undisputed. First, the ability of a particle to bind to a phosphocellulose column depends upon the initial salt concentration of the buffer in which the phos-phocellulose column is suspended. Prome-*1361ga’s expert, Dr. Mosbaugh, concurred with Roche’s evidence on that point. Second, the initial salt concentration of the Stoffel experiment (80 mM) differed significantly from that described in the Chien reference (10 mM). Because the Stoffel experiment was conducted under conditions very different from those described in the Chien reference, it is not clear that the results of the Stoffel experiment were pertinent to the inventors’ characterization of the prior art enzyme. Because the district court improperly concluded as matter of law that the Stoffel experiment was material, the portion of the district court’s summary judgment order relating to the materiality of the Stoffel experiment cannot be sustained.
We also must reverse the district court’s finding of intent with respect to the Stoffel experiment. Following trial, the court found that the inventors intended to deceive the PTO when they failed to disclose the results of the Stoffel experiment to the examiner. Roche characterizes that finding as clearly erroneous, and we agree. The evidence does not demonstrate that the inventors intended to deceive the PTO by suppressing the results of the Stoffel experiment. Instead, it shows that they believed that the initial salt concentration in such an experiment was critical to the capacity of the fragment to bind to the column and that the low initial salt concentration used in the Chien experiment rendered the Stoffel experiment not comparable and the results of the Stoffel experiment not pertinent to patentability.
The evidence showed, for example, that when one of Dr. Gelfand’s colleagues had difficulty with an enzyme sticking to the column, he called Dr. Gelfand, who concluded that it was possible the colleague had encountered a fragment and suggested reducing the salt concentration to around 40 mM. The colleague stated that “[t]he suggestion from Dr. Gelfand was that the salt concentration of the load in this column was probably too high. And in order to get that material to stick to the column, one needed to lower the salt concentration by dialyzing.” Similarly, Ms. Stoffel testified that experiments showed that the fragment did not bind when it was loaded at a salt concentration of 80 mM, but that “it stuck but it eluted earlier in the gradient” at a salt concentration of 10 mM — the concentration in Chien’s experiment. When the inventors noted that the proteolyzed enzyme did not bind to phos-phocellulose, they included a reference to the restriction “at 80 mM salt,” indicating that they considered the salt concentration significant to the failure to bind. Thus, the record reflects that the enzyme’s ability to bind to the column is dependent on the initial salt concentration and that the inventors knew of that dependence and deemed it significant. This evidence as to the inventors’ state of mind stands as un-rebutted proof that they did not think the results of the Stoffel experiment had any bearing on whether Chien had isolated a fragment of the Taq enzyme. Accordingly, we conclude that the district court committed clear error in finding clear and convincing evidence of intent to deceive in the inventors’ failure to disclose the Stoffel experiment.
2. Size Exclusion Chromatography
The district court also ruled that the inventors improperly withheld four pieces of information that suggested that the technique Chien used to determine molecular weight — size exclusion chromatography — was inaccurate and produced results underestimating the true molecular weight of Chien’s Taq. In the district court’s view, that information would have weakened the inventors’ argument that the claimed enzyme was distinct from the en*1362zyme of the prior art, as evidenced by a difference in molecular weight.
In the size exclusion chromatography procedure in question, a sample of the protein whose weight is to be estimated is introduced into a column filled with porous beads (sometimes referred to as a “matrix”). A buffer solution is run through the column. The proteins in the sample that are larger than the pore size of the beads will flow around the beads and out the bottom of the column. The smaller proteins will enter the pores between the beads, thus retarding their progress through the column. As a result, the speed at which a protein “elutes,” or moves through the column, corresponds to its molecular weight. The faster a protein moves through the column, the greater its molecular weight. If, however, a protein that otherwise would pass quickly through the matrix because of its large size has a chemical affinity for the matrix, it may move through the column more slowly than would be expected, resulting in an underestimation of its molecular weight.
According to the distinct court, the inventors had information suggesting that the Taq enzyme would have bound to the matrix used by Chien in her size exclusion procedure. The district court therefore concluded that the inventors knew that Chien’s molecular weight value was potentially too low, and that the difference between the claimed and prior art enzymes was less than the inventors asserted.
The information to which the district court pointed consisted of (1) the results of a purification experiment by Ms. Stoffel; (2) a sizing experiment by Dr. Drummond, another Cetus employee; (3) data concerning Taq’s hydrophobicity (its tendency to avoid interacting with water); and (4) an internal memorandum by Mr. Raymond, a Cetus employee. Roche argues that the information marshaled by the district court does not constitute evidence that Taq would have bound to the matrix used by Chien in her sizing experiment and that the district court improperly ruled in Promega’s favor on this issue. We agree. None of the information cited by the district court tends to prove that Taq would have bound to the particular type of matrix used by Chien, a product known as Sephadex G-100.
The experiments conducted by Dr. Drummond and Ms. Stoffel used types of gels (Zorbax and Ultrogel) that differed from the type of gel used by Chien. Promega’s expert on this subject, Dr. Burgess, conceded that the phenomenon of a protein binding to a matrix is dependent upon the specific chemistry between the matrix and the protein in question. Promega introduced no evidence that the experiments that were conducted with Zorbax and Ultrogel shed any light on the manner in which Sephadex and Taq interact. The evidence therefore did not show that the experiments indicated that Chien’s Taq would have bound to the matrix and produced inaccurate results. Accordingly, the district court improperly held that the failure to disclose those experiments was a material omission.
The district court also ruled that the inventors’ knowledge of Taq’s hydrophobicity was material to determining whether Chien’s sizing experiment was accurate. Roche contests that ruling, claiming that Taq’s hydrophobicity is not probative of whether it would interact with a matrix composed of Sephadex.
It is undisputed that a protein’s interaction with a particular sizing matrix depends not only on the properties of the protein, but also on those of the matrix itself. In reaching the conclusion that Taq’s hydrophobicity would make it likely that it would have bound to Sephadex, the district court did not consider the chemical properties of Sephadex. That omission is *1363critical. The only evidence adduced at trial on that point was from Roche’s expert Dr. Chamberlin, who testified that the chemical composition of Sephadex would make it unlikely to bind to a hydrophobic enzyme such as Taq, because Sephadex is hydrophilic rather than hydrophobic. Thus, the record does not support the district court’s conclusion that Taq’s hydrophobic nature makes it likely that Chien’s sizing experiment produced inaccurate molecular weight estimates, because there is no proof that Taq is likely to bind to a matrix composed of Sephadex G-100. The court’s finding on that issue is clearly erroneous.
Finally, the district court ruled on summary judgment that the Raymond memorandum was material. That memorandum indicates that Taq “migrates differently on sorbax [sic: Zorbax] or other sizing columns” and that a different technique, SDS — PAGE, is needed to get an accurate determination of molecular weight. Again, however, the Raymond memorandum does not indicate whether Taq would interact with Sephadex G-100. The possibility that the Raymond memorandum was referring to Sephadex when it mentioned “other sizing columns” is remote, because there is no evidence that Cetus ever conducted any sizing experiments with that type of gel. Furthermore, Roche contends that because Sephadex has a significantly different chemistry than the types of gels Cetus used, extrapolating from the Raymond memorandum in the manner that the district court did is improper. We conclude that, at a minimum, there is a genuine dispute of material fact on this issue, and we therefore hold that it was inappropriate for the district court to resolve this question on summary judgment. In sum, we reverse the district court’s inequitable conduct ruling to the extent that it relies on representations and omissions regarding the respective molecular weights of the prior art and the claimed enzymes.
B. Example VI
Following trial, the district court found that the inventors committed inequitable conduct by erroneously stating that they had performed a particular purification protocol and had obtained certain results. Example VI is a procedure described in the '509 application for repeatedly refining, or fractionating, a bacterial culture of the Taq enzyme through a variety of techniques, including column chromatography. According to Example VI, the last of these “fractions” is the purified version of the claimed enzyme. Example VI concludes:
Active fractions with no detectable nuclease(s) were pooled and run on a silver stained SDS-page mini gel. The results show a single 88 kd band with a specific activity of 250,000 units/mg.
This specific activity is more than an order of magnitude higher than that claimed for the previously isolated Taq polymerase and is at least an order of magnitude higher than that for E. coli polymerase I.
'818 patent, col. 41, ll. 12-20. The inventors further stated that “[t]he Taq polymerase purified as described above in Example VI was found to be free of any contaminating Taq endonuclease and exo-nuclease activities.” '818 patent, col. 41, 11. 23-25.
1. Misrepresentations
Example VI is written in the past tense. The inventors state, for example, that a certain quantity of cells “were resuspend-ed in 75 ml of a buffer,” that the cells “were lysed in a French press,” '818 patent, col. 39, ll. 3-7, after which 300 ml of Tris-EDTA “were added.” Each step of the example, over more than two columns *1364of the patent, is described in the same fashion, using the past tense. Indeed, the past tense is used to describe the steps of Example VI on more than 75 occasions. E.g., “These steps were repeated and the protein suspension was dialyzed extensively”; “The polymerase activity was assayed”; “An assay ... was performed”; “Active fractions with no detectable nuclease(s) were pooled, and run on a silver stained SDSPAGE mini gel.” '818 patent, col. 39, ll. 61-62; col. 41, ll. 3-4, 6-14. From the language used, a reader of the patent would conclude that the protocol was performed and that the following results were actually achieved: (1) the refined enzyme possessed “single-band purity”; (2) the purified enzyme was free from nuclease activity; (3) the enzyme had a specific activity of approximately 250,000 units/mg; and (4) the specific activity of the enzyme was at least ten times that of the prior art enzyme.
As a threshold matter, Roche suggests that the description of Example VI in the past tense is not actually a misrepresentation, and therefore that we need not inquire into the issues of intent and materiality. The district court, however, specifically found that Example VI had not been performed as written, and in light of the testimony adduced at trial, that finding cannot be held to be clearly erroneous.
Dr. Gelfand admitted that Example VI was never performed as described:
Q. Well, if you thought this was the best way, when did you ever do Example 6 from start to finish exactly as written? You never have done Example 6 from start to finish as written, have you, ever?
A. Susanne Stoffel and I, that is correct, Susanne Stoffel and I never carried out Example 6 from start to finish as laid out.
Ms. Stoffel’s testimony was similar. Since Example VI was never performed, the inventors never obtained the described results.
Roche notes that the last sentence in the description of the procedure purportedly used in Example VI is in the present tense (“The results show a single 88 kd band with a specific activity of 250,000 units/ mg.”). Because that sentence is in the present tense, Roche argues, a reader would understand that Example VI was a prophetic example, and was not actually performed. That argument is entirely unpersuasive. After having described in great detail the process purportedly used in the experiment and the results obtained — all in the past tense — the use of the present tense in the single quoted sentence would be understood as a present characterization of results actually obtained in the past, not as the inventors’ prediction of what such an experiment would show if it were performed.
Roche next suggests that the use of the past tense in Example VI is not really a misrepresentation since all of the steps in that procedure were purportedly part of two procedures that Roche did perform, Prep 3 and Prep 4. Specifically, Dr. Gel-fand stated that four of the five column steps of Example VI were completed in Prep 4 and the remaining step was performed at some point in Prep 3.
The district court, however, rejected the argument that Example VI was in effect performed through the combination of Prep 3 and Prep 4. The court noted that Gelfand and Stoffel had “combined steps from [Prep 3 and Prep 4] to arrive at Example VI,” but the court made clear that it did not regard the combination of steps from those two preparations as constituting the performance of Example VI. In fact, the court explained that the representation “that Example VI yielded a sin*1365gle 88 kd band on an SDS PAGE mini-gel was necessarily a misstatement because the inventors had not, in fact, performed Example VI of the patent.”
While Promega’s expert witness, Dr. Linn, conceded that the inventors did each of the column steps at some point, he also noted that they were not done in the order specified in Example VI. Moreover, Dr. Mosbaugh testified that the order of the steps is important and that the addition or deletion of a step would affect the collection of proteins, a significant result. He stated that “it’s not just a matter of doing five steps. It’s important what those five steps are, the order in which they’re carried out.” Dr. Mosbaugh examined the details of the steps and noted that there were many differences in how the two final columns in Prep 4 were “handled and evaluated relative to what the patent says” in Example VI. When asked about the predictive value of Prep 3 and 4 with regard to Example VI, Dr. Mosbaugh stated that one “cannot fuse these together and get any prediction as to what the outcome would be.” He concluded that “they didn’t do Example 6 as reported in the patent; [ ] what they did do, that is [Prep] 3 and [Prep] 4, will not get you, in my opinion, to Example 6.” Roche offered no evidence contradicting the importance of the order of the steps, the differences within the steps, and the conclusion that the results of Prep 3 and Prep 4 would not provide an accurate representation of the results of a procedure conducted according to Example VI. Accordingly, although Roche attempts to convert the district court’s finding that the inventors generated Example VI by combining steps from Prep 3 and Prep 4 into a finding that by performing Prep 3 and Prep 4 they had in effect performed Example VI, that effort runs headlong into the district court’s repeated findings that Example VI was never performed. And, as noted, that finding was supported by evidence and cannot be deemed clearly erroneous.
Nor is there force to Roche’s argument that the results reported in Example VI are consistent with the results of other experiments that had actually been performed. With respect to the assertion in Example VI that the inventors derived an enzyme exhibiting single-band purity, Dr. Mosbaugh stated that even improved protocols for purity failed to achieve single-band purity. He also noted that when outside contractors performed Example VI, they encountered multiple bands. Roche’s expert witness, Dr. Chamberlin, testified that he did not see a gel from either Prep 3 or Prep 4 with only a single band. Additionally, while Ms. Stoffel asserted that she achieved a single band in Prep 4, Dr. Gelfand conceded that in Prep 4 they obtained a major band, but that “it was not the only band. There were some other very faint bands.” Based on that testimony, the district court permissibly concluded that the representation that Example VI yielded a single band was a significant deviation from the information known to the inventors.
The district court reached a similar conclusion with respect to the assertion in the specification that Example VI resulted in a nuclease-free preparation. In particular, the court found that the representations that “[a]ctive fractions with no detectable nuclease(s) were pooled and run on a silver stained SDS PAGE mini-gel” and that “[t]he Taq polymerase purified as described above in Example VI was found to be free of any contaminating Taq endonuclease and exonuelease activities” were false. The court noted that while the inventors relied on the results of Prep 4 to support those claims, Ms. Stoffel and Roche’s expert admitted that the enzyme that resulted from that procedure was not nuclease-free. Roche’s expert^ Dr. Cham-*1366berlin, testified that there was detectable endonuclease activity. Ms. Stoffel testified that Prep 4 contained only a “very small” or “minimal” amount of nucleases, but Promega’s expert witness Dr. Linn characterized the contamination of that preparation as substantial. In light of this testimony, on which the district court relied, it was not clearly erroneous for the district court to conclude that the inventors did not achieve a nuclease-free polymerase in any experiment that they actually conducted.
The district court was also justified in concluding that the inventors’ assertion that they had achieved an enzyme with a specific activity of approximately 250,000 units/mg was false. Example VI specifically states that a “unit” is equal to “10 nM of product in 30 minutes.” The court found at the summary judgment stage, however, that the 250,000 value corresponded to a different definition of a “unit,” and it held that under the definition set forth in Example VI, the correct value should have been recited as 100,000 units/ mg. The district court noted that Dr. Gelfand gave conflicting ' accounts of the source of the specific activity value reported in Example VI. In a declaration submitted to the court, Dr. Gelfand asserted that the value was obtained according to the conditions recited in Example VI. In a prior declaration submitted to the PTO, however, Dr. Gelfand stated that the results of Example VI were obtained from an experiment based on the assay conditions listed in Example I (also in the '509 application). In reality, neither of those statements was true, for it appears that the 250,000 number was based on an experiment performed by Ms. Stoffel, in which she obtained a result of 248,000 units/mg.
The Stoffel experiment does not, however, prove that the specific activity result of Example VI is grounded in an actual experimental result. Although Dr. Gelfand testified that Ms. Stoffel’s result was 200,-000 units/mg “under the new unit definitions,” that testimony does not prove that the new unit definition was the same as the one used in Example VI, nor does it prove that the Stoffel experiment was conducted under conditions equivalent to those described in Example VI. Roche’s claim that another experiment measured a specific activity of approximately 390,000 units/mg also fails for similar reasons, because there is no proof that that experiment was performed under conditions equivalent to Example VI with a per milligram specific activity value approaching 250,000 “units,” as that term is defined in Example VI. Thus, Roche has not met its burden of demonstrating that the district court clearly erred in determining that the inventors’ stated specific activity result was a misrepresentation.
Finally, the evidence supports the court’s finding that the inventors’ comparison of the specific activity of the claimed and prior art enzymes was deceptive. The application stated that the specific activity of the claimed enzyme, as purified according to Example VI, was at least ten times that of the prior art enzyme. The inventors repeated that argument in the March 17, 1989, office action response as an alternative argument for patentability. Based on Dr. Mosbaugh’s testimony, however, the court found that the comparison was invalid because the two enzymes were not assayed under the same conditions. The court also found that the contrary testimony of Roche’s witness, Dr. Chamberlin, was not credible. There is no clear error in that determination.
2. Intent
Misrepresentations by themselves are not enough to render a patent unenforceable; the misrepresentations must be *1367intentional and they must be material to patentability. With regard to the element of intent, Roche has not demonstrated clear error in the district court’s finding. The inventors attested that all statements made in the '509 application were true. There was no suggestion by Roche that the use of the past tense in Example VI was an oversight — Dr. Gelfand admitted he understood that, at least in a scientific publication, the use of the past tense means that an experiment was actually performed. He provided no reasonable explanation as to why a different principle would apply in a patent application. Nor did Roche introduce any other evidence to explain why the past tense was used to describe an experiment that was not performed. Accordingly, the district court did not clearly err in determining that the inventors’ use of the past tense in Example VI was knowingly false.
Roche contends that Example VI was included in the application because it represented the best mode of practicing the invention, disclosure of which is required by 35 U.S.C. § 112. The best mode requirement, however, does not entitle the inventor to suggest that the best mode has been performed when it has not, and to report results that have not actually been observed.
Roche attaches great significance to the fact that the district court found that the inventors had a good faith belief that they had discovered a different enzyme than that described in the prior art, arguing that “[b]ecause one cannot intentionally deceive by representing what one honestly believes, the district court’s judgment cannot stand.” Roche misapprehends the import of that finding by the district court. The inventors may indeed have believed they had discovered a novel enzyme, but that belief does not permit them to make misrepresentations in seeking to persuade the examiner to issue a patent for that enzyme. Thus, the district court’s finding that the inventors had a good faith belief in the novelty of their invention is not incompatible with a finding of deceptive intent.
3. Materiality
As to the materiality element of inequitable conduct, Roche argues that because the '818 patent did not include claim limitations directed to purity, representations about purity made in the application cannot be material. Materiality, however, is not limited to matters reflected in the claims of a patent. See PerSeptive Biosystems, Inc. v. Pharmacia Biotech, Inc., 225 F.3d 1315, 1322, 56 USPQ2d 1001, 1006 (Fed.Cir.2000). Rather, information is material when there is a substantial likelihood that a reasonable examiner would consider it important in deciding whether to allow the application to issue. Id. at 1321, 56 USPQ2d at 1006. Moreover, the information at issue with respect to Example VI consisted of affirmative misrepresentations, not omissions, as was the case with respect to the issue of molecular weight. This court has held that affirmative misrepresentations by the patentee, in contrast to misleading omissions, are more likely to be regarded as material. Rohm & Haas Co. v. Crystal Chem. Co., 722 F.2d 1556, 1571, 220 USPQ 289, 300 (Fed.Cir.1983).
In the office action response, the inventors argued that even if the claimed enzyme was not distinct from the prior art enzyme, “[a]pplieants would still be entitled to a patent because the present preparations are far more pure ” than the prior art enzyme, citing the comparison of specific activity made in the patent application. Purity was therefore a prominent (albeit an alternative) argument in favor of patentability. Indeed, the prosecuting attorney testified that the argument regard*1368ing purity was made in order to allow amendments in the event that the examiner did not allow the existing claim. The district court based its finding of materiality on that ground: “Since Cetus argued that the patent could issue based on the asserted purity limitation, a reasonable examiner would have considered important information which indicated that Cetus had overstated the level of purity of the claimed enzyme.”
The court’s determination is not clearly erroneous. The fact that the examiner did not have to rely on the purity representations in issuing the patent is not inconsistent with a finding of materiality. See Merck & Co. v. Danbury Pharmacal, Inc., 873 F.2d 1418, 1421, 10 USPQ2d 1682, 1686 (Fed.Cir.1989) (rejecting a “but for” standard of materiality). Although the inventors’ statements regarding purity were not the principal focus of the office action response, they were clearly an important aspect of it. Under the circumstances, a reasonable examiner would have wanted to know that the patentability argument based on purity was unsupported by the experimental results cited by the inventors.1 Thus, the purity results reported in Example VI and the reference to them in the office action response were central enough to the prosecution to be “within a reasonable examiner’s realm of consideration.” Id.2
C. Comparison of Fidelity
The district court next focused on the inventors’ comparison of the fidelity and template dependency of the claimed enzyme and the prior art enzymes. A DNA polymerase is said to be “template dependent” if it has the ability to read the identity of a nucleotide in the template strand of DNA and then synthesize the appropriate complementary base pair. Fidelity relates to the accuracy of the DNA polymerase in adhering to the complementary base-pairing during synthesis. A polymerase exhibits a higher fidelity when it experiences fewer misincorporation events, *1369which are mistaken base-pairings. A template-dependent, high fidelity DNA polymerase will exhibit little or no activity when one of the four types of nucleotides is absent from the reaction mixture. The polymerase will incorporate nucleotides until it encounters a position on the template for which the corresponding complement is not available. At that point, the polymerase will effectively stop working because it cannot perform the synthesis operation correctly, causing experimental results of less than one hundred percent incorporation.
In the '509 application, the inventors posited that their claimed enzyme exhibited high fidelity while the prior art enzyme experienced misincorporation. The inventors stated that:
when only one or more nucleotide triphosphates were eliminated from a DNA polymerase assay reaction mixture, very little, if any, activity was observed using the enzyme herein, and the activity was consistent with the expected value, and with an enzyme exhibiting high fidelity. In contrast, the activity observed using the Kaledin et al. (supra) enzyme is not consistent with the expected value, and suggests misincorporation of nucleotide triphosphate(s).
'818 patent, col. 30, 11. 23-31. The inventors made a similar assertion in their March 17,1989, office action response:
The data in the ... [prior art references] would lead one of ordinary skill in the art to conclude that the enzymes described in these references are not suitable for template-directed in vitro DNA synthesis, because the enzymes have a rather substantial promiscuous ability to synthesize DNA on a natural DNA template in the absence of one of the four deoxynucleoside triphosphates.
The Taq polymerase of the present invention has an activity quite different from and superior to the activity of the polymerase described by [the prior art references]. On page 47, lines 6-12, of the specification, Applicants point out that the purified Taq polymerase of the invention has little or no activity in a DNA polymerase assay reaction mixture that does not contain one of the four deoxynucleoside triphosphates.
That assertion was made to support the contention that the claimed enzyme was different from the prior art enzyme cited by the examiner. The claim of greater fidelity was therefore intended to demonstrate, along with the purported difference in molecular weight, that the claimed enzyme was not the same as the enzyme described in the prior art references and accordingly was patentable.
While it appears that the applicants’ statements about the fidelity of their own enzyme were accurate, the district court found that the statements characterizing the activity of the prior art enzymes and comparing it to that of the claimed enzyme were inaccurate. That finding is not clearly erroneous. Promega’s expert witness Dr. Kunkel testified that the Kaledin and Chien experiments are “not indicative of an enzyme that is not suitable for template-directed in vitro synthesis,” but instead indicate just the opposite. Dr. Kunkel also addressed the applicants’ representations that the prior art enzyme had a substantial promiscuous ability, which in the field of fidelity measurements suggests misincorporation, and that the reported activity levels indicated misincorporation.3 *1370Dr. Kunkel explained how Kaledin’s and Chien’s levels of incorporation recorded in the absence of one of the nucleotides did not reflect misincorporation events, but rather that the enzyme was incorporating until the missing nucleotide was needed for a base-pairing. Based on an examination of the structure of the template used by Kaledin, Dr. Kunkel reported that the results that Kaledin obtained were as anticipated, given the nature of the experiment, and that the experiment in no way indicated misincorporation.
According to Dr. Kunkel, the differing results referred to by the inventors in the office action response related to differences in experimental conditions (specifically, differences in the DNA templates used), and not to differences in the properties of the enzymes. For example, Dr. Kunkel explained that the template used by Kaledin and Chien was a mixture of double- and single-stranded DNA with various short, single-stranded gaps and that the synthesis began at a different point in each molecule. Therefore, there were as few as 21 total incorporation events in one example. In a particular instance, Dr. Kunkel testified, a gap may not even require the missing nucleotide and would incorporate fully, adhering to the complementary base-pairing and not misincorporating. This would result in activity levels closer to 100%. In contrast, the template used by Dr. Gelfand is a long single-strand that the enzyme must copy with synthesis starting at the same point on all the molecules and approximately 7000 incorporation events. A low percentage of the total incorporation is likely then to occur before the missing nucleotide is needed, resulting in lower activity levels than in Kaledin, but not necessarily indicating a higher fidelity. Dr. Kunkel explained that the results illustrated that both of the polymerases were template-dependent, that Kaledin’s and Dr. Gel-fand’s results were as anticipated, and that the percent incorporation relative to the control is a function of the nature of the DNA. In addition, Dr. Kunkel described an experiment he conducted in which the same enzyme produced varying activity levels when different templates were used. According to Dr. Kunkel, that experiment demonstrated that the different activity levels found by Kaledin and by the inventors resulted from the use of different templates, “illustrating the well-known principles taught in the literature that the difference in incorporation values that one gets are a function of the DNA template that was used[,] not a function of the polymerase that was used.” Dr. Kunkel testified that misincorporation could not be an explanation for the activity levels that Ka-ledin and Chien were observing. Roche did not offer expert testimony to contradict Dr. Kunkel’s statements.
Roche responds to the district court’s analysis of this issue by arguing that the inventors and Dr. Kunkel simply differed over how to interpret the results of the experiments involving the prior art and claimed enzymes, and that a dispute of that sort cannot give rise to a finding of inequitable conduct. Roche, however, failed to make a persuasive showing that there was a legitimate difference of scientific opinion on that issue and that the district court therefore committed clear error in its findings.
In its brief, Roche argues that a portion of Dr. Kunkel’s cross-examination, together with a passage from the Chien reference itself, supports the claim that the comparison between the enzymes was sound. Roche argues that Chien explained that her activity level results were related to the nature of the enzyme, rather than to the template used. Chien stated that the incorporation she observed with one nucleotide omitted is “higher than those reported for most bacterial polymerases but *1371is similar to certain mammalian DNA polymerases. The incorporation probably represents the addition of two or three dTTP’s to complementary ends.” However, Dr. Kunkel stated that Chien’s use of the term “complementary” implies correct matching of base-pairs, rather than misin-corporation. Thus, Chien fails to provide the requisite evidence that the district court committed clear error in rejecting Dr. Gelfand’s statement that the activity suggests misincorporation. Roche’s argument and citation to Chien fall far short of rebutting Dr. Kunkel’s detailed testimony on which the district court relied. The court therefore did not clearly err in finding that the inventors misrepresented the prior art enzyme and its activity as compared to that of the claimed enzyme.4
We likewise uphold the district court’s finding of materiality with respect to the fidelity issue. The statement in the office action response was made in an effort to support the inventors’ primary argument for patentability—that the claimed enzyme was different from the prior art enzyme. The most prominent evidence cited by the inventors in this regard was a difference in molecular weight, but the alleged difference in fidelity was also argued at some length. Given the prominence of that argument in the office action response, the district court did not clearly err in concluding that the misrepresentation made as part of that argument was material.
The district court found that the statements about fidelity were intentionally deceptive, based primarily on testimony from Dr. Kunkel, who worked with Dr. Gelfand and Cetus. Dr. Kunkel stated that he was aware of Dr. Gelfand’s knowledge of template dependency and fidelity and that Dr. Gelfand could not possibly believe that the experiments indicated that the prior art enzyme was promiscuous and not suitable for template-directed synthesis. Dr. Kunkel testified that he did not believe that a scientist aware of the information in the Cetus records could truthfully make the statements found in the specification and office action response and that Dr. Gelfand was knowledgeable about that information. Stating that the experiment done in the prior art had nothing to do with fidelity or misincorporation and did not indicate that the prior art enzyme was not suitable for template-directed in vitro synthesis, Dr. Kunkel noted that such information was “well known to persons who studied polymerases in the '70’s and '80’s.” In Dr. Kunkel’s words, “with respect to fidelity, the issue that I was just speaking to, I feel very strongly that the that Dr. Gelfand knew better.”
Roche asserts that Dr. Kunkel’s testimony is not probative of intent and that the finding of intent cannot stand because neither the inventors nor the prosecuting attorney were asked about that issue. Intent, however, is typically proved inferentially, Molins PLC v. Textron, Inc., 48 F.3d 1172, 1180, 33 USPQ2d 1823, 1828 (Fed.Cir.1995), and a finding of intent does not require a confession from the stand by the inventor or the prosecuting attorney. The district court’s reliance on Dr. Kunk-el’s testimony was entirely permissible. Dr. Kunkel is an expert in enzyme fidelity and was invited by Dr. Gelfand to visit Cetus, where he presented a seminar on the fidelity of Taq. Dr. Kunkel had discussions with Dr. Gelfand on that subject both at Cetus and at conferences where Dr. Kunkel presented. Accordingly, he was well positioned to testify as to Dr. Gel-fand’s knowledge of fidelity and misincor-*1372poration. While Dr. Kunkel did not testify that he and Dr. Gelfand spoke about the specific subject of the impact of a template on the percentage incorporation occurring in the absence of one of the nucleotides, it was permissible for the district court to infer that Dr. Gelfand understood that relationship on the record before it. Dr. Kunkel’s testimony provided a sufficient basis for the district court’s conclusion that Dr. Gelfand had sufficient information to understand the import of the statements made in the specification and the office action response. The district court did not clearly err in finding the requisite intent.
We note that Roche did not put forward evidence to contradict Dr. Kunkel as to whether the inventors’ interpretation was reasonable, even though inaccurate. For example, Roche did not put forward testimony by another scientist, in a position similar to Dr. Gelfand, that a scientist studying polymerases could have misunderstood the data to indicate that the prior art enzymes were misincorporating and exhibiting a lower fidelity than the claimed enzyme. In particular, Dr. Gelfand did not testify about his knowledge of fidelity or that he believed the statements made in the specification and in the office action response. We therefore have no eviden-tiary basis from which to conclude that it was clearly erroneous for the district court to find that the assertions regarding fidelity were not legitimate or reasonable, and that the inventors therefore intentionally sought to deceive the PTO.
Ill
As we have explained, an important step in the judicial resolution of inequitable conduct claims is for the court to determine whether the material misrepresentations or omissions in question are sufficiently serious in light of the evidence of intent to deceive, under all the circumstances, to warrant the severe sanction of holding the patent unenforceable. That determination is committed to the trial court’s discretion. In this case, the trial court did not expressly address this step in the inequitable conduct analysis. For that reason, and because we have not upheld all of the grounds on which the court found inequitable conduct in this case, we vacate the court’s order of unenforceability and remand the case to the district court for it to determine, in the exercise of its judgment, whether under all the circumstances, the incidents of inequitable conduct that we have sustained are such as to justify the sanction of rendering the '818 patent unenforceable.
Each party shall bear its own costs for this appeal.
VACATED and REMANDED.
. Regents of the University of California v. Eli Lilly & Co., 119 F.3d 1559, 43 USPQ2d 1398 (Fed.Cir.1997), relied on by the dissent, is inapposite, because the misidentification of a plasmid in the patent examples did not make the examples inoperative; rather, the procedures described in the examples "worked to yield the intended results irrespective of" whether the actual plasmid was used, making the "misidentification of the plasmid ... not material to patentability." Id. at 1570-71, 43 USPQ2d at 1408. In this case, not only did the applicants fail to practice the example as written, but they also failed to achieve the stated results. The dissent’s reliance on Atlas Powder Co. v. E.I. du Pont De Nemours & Co., 750 F.2d 1569, 224 USPQ 409 (Fed.Cir.1984), is also misplaced. In that case, the examples were written in the present tense to conform with the PTO requirements on prophetic examples, and the appellant’s claim was that the patentee should have informed the patent examiner that the examples written in the present tense were prophetic. We held that it was not clear error for the district court to find no materiality or intent to deceive under those circumstances. Not only were those facts— use of the present tense when it was proper to do so — wholly different from the facts of this case, but in Atlas Powder we applied the "clearly erroneous” standard of review to uphold the district court's findings, as opposed to rejecting the district court’s findings as the dissent would have us do here.
. Under the current standard for materiality, there would be no question that the inventors' representations would be material. See 37 C.F.R. § 1.56(b)(2) (2002) ("[information is material to patentability when ... [i]t refutes, or is inconsistent with, a position the applicant takes in ... [ajsserting an argument of patentability.”). Although the current standard was not in effect at the time of the prosecution of this patent, the new standard was not intended to constitute a significant substantive break with the previous standard. See 57 Fed.Reg. 2021, 2023 (Jan. 17, 1992) (explaining that the amendment to § 1.56 was intended to clarify the lack of certainty in the previous materiality standard).
. Roche contends that the language at issue is merely a statement of belief, which cannot constitute a material misrepresentation. We disagree. The applicants did not in any way indicate that they were simply conveying their personal beliefs; instead, the language used by the applicants indicated that they were offering the representations as statements of fact.
. We do not suggest that an applicant is required to repeat the experiments of the prior art; an applicant, however, may not knowingly misrepresent those experiments or their results.