OPINION OF THE COURT
ADAMS, Circuit Judge.
In United States v. Baynes (Appeal of Trice), 622 F.2d 66 (3d Cir. 1980), this Court ordered that an evidentiary hearing be held with respect to the habeas petition brought by Gregory Trice, who claimed he had not received the effective assistance of counsel at his trial. On remand, after conducting such a hearing, the district court determined that Trice’s constitutional right to effective assistance had been infringed, but held that no prejudice resulted to the petitioner therefrom. We agree that Trice’s sixth amendment rights were violated by his trial counsel’s admitted failure to investigate potentially exculpatory information. Unlike the district court, however, we cannot conclude that Trice was not prejudiced by his counsel’s shortcomings. Accordingly, the judgment of the district court will be reversed.
I
Appellant Trice, along with seven co-defendants, was indicted in 1974 for conspiracy to distribute and possess heroin. In February 1975, he was convicted after a jury trial and sentenced to a prison term of fifteen years, which he immediately began serving. That conviction was affirmed without opinion by this Court, United States v. Baynes, 547 F.2d 1164 (3d Cir. 1976).
Subsequently Trice filed a petition for a writ of habeas corpus, alleging that he had been inadequately assisted by counsel at trial. The district court denied this application without a hearing. On appeal, this Court reversed and remanded for an evidentiary proceeding pursuant to 28 U.S.C. § 2255. United States v. Baynes, 622 F.2d 66 (3d Cir. 1980) (hereafter cited as Baynes).
According to this Court’s opinion in Baynes, although “substantial evidence” was proffered at trial against each of his co-defendants, “the Government’s entire case against Trice was predicated on an electronically intercepted telephone conversation of merely twelve words allegedly involving Trice and implicating him in the drug conspiracy.” 622 F.2d at 67. Prior to trial, the Government had obtained from Trice a voice exemplar; however, that exemplar was never introduced into evidence by the prosecution, a fact which this Court labeled “surprising.” Id. On his habeas appeal, Trice contended that, despite his repeated urgings, his defense counsel had made no attempt to listen to the exemplar and compare it with the Government’s intercepted recording. Trice insisted that it was not his voice on the telephone recording, and that any diligent investigation by his lawyer would have revealed this fact and led to Trice’s exoneration by the jury. Rather than conduct such an investigation, Trice’s counsel did nothing with respect to the exemplar, choosing instead to cross-examine the officer at trial who identified the [663]*663voice on the intercepted recording as belonging to Trice.1 Id. at 68.
The Court in Baynes accepted as true the allegations contained in Trice’s petition, as required by the habeas statute. After reviewing this Circuit’s standard for measuring inadequate assistance claims, discussed infra, the Court concluded that Trice had made out a ease sufficient to warrant an evidentiary hearing. The panel expressed no opinion, however, whether Trice was entitled to collateral relief:
That determination can only be made after the district court makes a full inquiry into the facts surrounding trial counsel’s alleged failure to investigate and utilize the voice exemplar. Such an inquiry may well require comparison of the voice exemplar with the intercepted telephone conversation to determine whether counsel’s failure to explore the exemplar issue prejudiced Trice’s ease.
Id. at 69-70.
At the hearing on remand, Trice testified that he repeatedly had urged his attorney to listen to the voice exemplar, “because I was certain that the conversation that was allegedly attributed to me in this conspiracy was not, in fact, me at all, and I felt that the exemplar would bring . . . this to light.” Transcript of proceeding (July 31, 1980), Appendix at 68-69. Trice’s trial counsel, who also testified at the habeas hearing, admitted that Trice “consistently” had indicated that his was not the voice on the intercepted recording; in fact, Trice had rejected a plea bargain arrangement of “four or five years” imprisonment, contending that he was innocent of the charges. Id. at 36, 39. Trial counsel also testified that, although he had accompanied his client when the voice exemplar originally was made, he had never listened to it again, and thus had never compared the exemplar with the intercepted recording. Id. at 35. When asked to explain why he had not made use of the voice exemplar in preparing Trice’s defense, the attorney observed: “Frankly, I was fearful of using it. I didn’t know what it would say and if it — if it were close, as a trial tactic [it] would have been stupid, I thought, in my humble opinion.” Id. at 48. Additionally, counsel disclosed that, during the trial itself, when the intercepted conversation allegedly involving Trice was being played in court, counsel was informed by one of Trice’s co-defendants that the taped voice in fact belonged to him (the co-defendant) and not to Trice.2
Relying primarily on the above, the district court concluded that the attorney who represented Trice at trial had been constitutionally ineffective, in that he had failed to conduct an investigation designed to reveal whether the exemplar and the intercepted recording contained the same voice. Despite this finding, the court held that Trice was not entitled to habeas relief because he [664]*664was unable to demonstrate that his attorney’s negligence had prejudiced his defense.
First, the district court concluded that Trice was not prejudiced by his trial counsel’s failure to procure a scientific comparison of the intercepted recording and the voice exemplar. Over Trice’s objection, see Appendix at 197-204, and despite the Government’s expressed reservation that “the technical uncertainties concerning the present practice of voice identification are so great as to require that forensic applications be approached with great caution,” see id. at 205-06, the district judge ordered that a spectrographic analysis of the two recordings be prepared by a government expert. That expert subsequently concluded, in a written report to the court, that the voices on the intercepted recording and the exemplar belonged to “one and the same person.”3 The district court reasoned therefrom that because “[t]he results of the spectrograph analysis suggest that Trice was a speaker in the subject incriminating telephone conversation,” his trial attorney’s neglect in not obtaining a spectrographic assessment could not be deemed prejudicial, inasmuch as “a scientific comparison by trial counsel would not have yielded evidence of an exculpatory nature or of a nature possibly altering the jury’s decision.”
Neither, the district court continued, could Trice demonstrate prejudice resulting from his trial counsel’s failure to conduct an aural comparison of the exemplar with the intercepted telephone recording. At the habeas hearing, Trice established that the two recordings contained varying pronunciations of three different words.4 Additionally, the Government witness who testified that, in his opinion, the same person was speaking in both recordings, on cross-examination not only admitted that he had not detected the three pronunciation discrepancies, but also conceded that it was “absolutely” possible for someone else to conclude, after aurally comparing the two recordings, that the voices contained thereon belonged to different persons.5 Despite this evidence and testimony, however, the district court was unpersuaded that trial counsel’s failure to discover and exploit the pronunciation differences on the two tapes [665]*665prejudiced Trice’s defense. To begin with, the district court observed that even though nothing in the record “suggest[ed] that Trice modified his speech either intentionally or unintentionally” in making the voice exemplar, it was a matter of “common sense” that “this in fact may have occurred,” considering that “the motivation and opportunity to give a deceitful exemplar is easily discernable.” More to the point, the district court dismissed the three pronunciation differences uncovered at the habeas proceeding as “de minimis,” in view of the fact that more than 400 words, some of them “difficult” and “capable of more divergent pronunciation” were uttered identically on the two recordings. In addition, the name of the police officer (“D’Amico”) was mispronounced consistently in the voice exemplar and the recorded telephone conversation. The district court therefore “found” that even had the jury been made aware of the pronunciation variances,
[its] verdict would have remained unaltered, and thus Trice was not prejudiced. The difficulty of speculating upon the possible effect of this newly-discovered evidence pales in comparison to the Court’s recollection of the entire trial and the conduct of all participants .... The Court’s conclusion that the jury’s verdict would not have been altered is unequivoeable [sic],
(citation omitted).
Finally, during a portion of the intercepted conversation, the parties discussed a contemplated visit, apparently scheduled for some time in May 1974, to a certain prisoner at the State Correctional Institution at Graterford, Pennsylvania. Eecords furnished by prison officials for examination during the habeas proceedings, however, indicated that, although the prisoner had received five visitors during that month, none of the visitors identified himself as Trice. The district court concluded that trial counsel’s failure to obtain the prison visitation records and to bring the contents of the records to the attention of the jury was not prejudicial to Trice’s defense: “no words were exchanged [in the conversation] concerning a specific date of visitation [to the prison], and this vagueness, coupled with the realization that even the best laid plans are discarded, prohibits the Court from attaching undue weight to its possible effect on the jury.”
Deeming the “allegations” of prejudice “too remote” to warrant relief, the district court denied Trice’s habeas petition. This appeal followed.
II
We first review the district court’s finding that Trice’s trial counsel performed ineffectively in failing to investigate what Baynes, 622 F.2d at 69, referred to as “a critical source of potentially exculpatory evidence” — namely, the voice exemplar recorded by the defendant prior to trial.6
In Moore v. United States, 432 F.2d 730 (3d Cir. 1970), this Court, sitting in banc, set forth the standard to be employed when evaluating ineffective assistance allegations. We held that an attorney, not unlike any other professional, is required to exercise “the customary skill and knowledge which normally prevails at the time and place.” Id. at 736. Necessarily, application of this standard entails a careful inquiry into the particular circumstances surrounding each case:
The defense attorney’s function consists, in large part, of the application of professional judgment to an infinite variety of decisions in the development and prosecution of the case. A determination whether any given action or omission by counsel amounted to ineffective assistance cannot be divorced from consideration of the peculiar facts and circumstances that influenced counsel’s judgment. In this fact-laden atmosphere, categorical rules are not appropriate.
[666]*666United States v. Decoster, 624 F.2d 196, 203 (D.C.Cir.1979) (in banc) (plurality opinion).
The Government argues that the facts elicited at the habeas proceeding below do not make out a case of ineffective assistance of counsel. First, according to the Government, counsel’s “failure” to refer to or make use of the voice exemplar can be explained as a matter of considered trial strategy. At the habeas hearing, Trice’s attorney testified that he simply was “fearful” of using the exemplar, Appendix at 48 — that is, presumably afraid of having the jury hear Trice read the same incriminating words he allegedly spoke during the intercepted telephone conversation. Additionally, the jury could well have wondered, had the exemplar been introduced and made a part of the defense’s case-in-chief, why Trice himself did not take the stand and testify, so that a more exact comparison might be made of his actual voice and the intercepted recording. Moreover, undue reliance upon the exemplar might have undermined counsel’s reasoned trial tactic of “lying low”: because the evidence against each of Trice’s co-defendants was much more substantial, counsel may have concluded that introducing the exemplar would focus unnecessary attention on his client, thereby lessening his chance of acquittal. Second, the Government observes that counsel conducted a vigorous and searching cross-examination, not only of the officer who identified Trice’s voice at trial, but also of other prosecution witnesses. The attorney’s “well prepared, alert, and aggressive performance throughout the trial [makes it] impossible to say that his overall performance on [Trice’s] behalf fell below the customary level of skill and knowledge of defense counsel.” Brief for Appellee at 17. Third, the Government notes that Trice’s trial counsel talked with the defendant on a number of occasions prior to trial — and consequently became familiar with the defendant’s voice “both over the telephone and in person.” Appendix at 47b. Trial counsel had no 'reason to compare the voice exemplar with the intercepted tape, the Government contends, inasmuch as counsel must have concluded, based on his own knowledge of Trice’s voice, that the intercepted speaker in fact was his client. The Government concludes that these three considerations, taken as a whole, demonstrate that Trice’s sixth amendment right to the effective assistance of counsel was not abridged by his trial attorney.
Each of the arguments proffered by the Government in defense of trial counsel’s performance, however, is fatally flawed. None goes to the central question posited by this Court in Baynes, namely, whether trial counsel’s acknowledged failure to investigate the exemplar evidence in the first place can be justified. We note that the first proposition advanced here by the Government — that Trice’s attorney refrained from discussing the exemplar as a matter of strategy — previously was tendered to the Baynes panel in an effort to counter Trice’s request for an evidentiary hearing. See 622 F.2d at 68. The Baynes panel made it quite clear, however, that assertions of this sort, in effect, missed the point: “Although the decision whether or not to utilize a particular item of evidence may be a matter of trial strategy within the acceptable bounds of trial counsel’s discretion, we believe the failure to investigate a critical source of potentially exculpatory evidence may present a case of constitutionally defective representation.” Id. at 69 (emphasis added). In this case, an avenue that conceivably might have led to the exoneration of the defendant was not explored by trial counsel: no attempt was made to compare the voice exemplar with the intercepted tape. To be sure, had such a comparison been made, Trice’s attorney might well have decided as a matter of trial strategy not to refer to the voice exemplar at trial. He could have arrived at a reasoned judgment, for example, that the voices contained on the two tapes were sufficiently similar that the jury might actually consider the exemplar to be inculpatory evidence against Trice; similarly, he might have determined that introduction of the exemplar would conflict with his overall desire to “lie low” and hope for the best for [667]*667his client.7 Such a decision on the part of trial counsel properly could have been made, however, only after a careful and comprehensive comparison of the two recordings had been conducted.8
Similarly, the fact that Trice’s counsel may have performed impressively at trial would not excuse a failure to investigate a defense that may have led to the complete exoneration of his client. In Moore, we rejected the suggestion that an attorney’s adequate trial presentation could make up for insufficient pretrial preparation:
[Representation involves more than the courtroom conduct of the advocate. The exercise of the utmost skill during the trial is not enough if counsel has neglected the necessary investigation and preparation of the case or failed to interview essential witnesses or to arrange for their attendance. Such omissions, of course, will rarely be visible on the surface of the trial, and to that extent the impression of a trial judge regarding the skill and ability of counsel will be incomplete.
432 F.2d at 739 (footnote omitted); see also Wolfs v. Britton, 509 F.2d 304, 309 (8th Cir. 1975) (“we cannot minimize the fact that effective assistance refers not only to forensic skills but to painstaking investigation in preparation for trial”); Rummel v. Estelle, 498 F.Supp. 793, 797 (W.D.Tex.1980) (“competent cross-examination does not take the place of affirmative defense evidence built upon thorough investigation and preparation”). Ultimately, as the district court recognized, the crucial question is whether Trice’s counsel should have investigated the exemplar evidence; if that failure to investigate cannot be explained, a finding of ineffective assistance would appear warranted, however nobly counsel performed at trial.
Finally, the fact that Trice’s attorney was acquainted with the sound of his client’s voice cannot serve as a bar to the allegation of ineffective assistance raised in this case. It is well settled that an attorney is obligated to examine potentially exonerating evidence even if the accused has admitted his guilt or stated his desire to plead guilty. See American Bar Association Standards, The Defense Function § 4.1 (1971); Gaines v. Hopper, 430 F.Supp. 1173, 1176-79 (M.D.Ga.1977), aff’d, 575 F.2d 1147 (5th Cir. 1978). This being the general rule, Trice’s attorney cannot be excused from examining the exemplar evidence under the circumstances here simply because he was of the opinion that it was indeed his client’s voice on the telephone conversation intercepted by the authorities. In the final analysis, it is the responsibility of the jury, and not the defense counsel, to draw conclusions of this sort. See generally Mitchell, The Ethics of the Criminal Defense Attorney— New Answers to Old Questions, 32 Stan.L.Rev. 293, passim & 298 n.14 (1980).
Having found the Government’s arguments attempting to justify trial counsel’s performance to be unpersuasive, we conclude, as did the district court, that the actions taken (or, more precisely, not taken) by Trice’s trial counsel with respect to investigation of the voice exemplar fell below the requisite knowledge and skill required to be exercised under Moore. It will be [668]*668recalled that the Government’s entire case against Trice rested on one short telephone conversation. Obviously, if counsel could have established, by whatever means, that it was not Trice’s voice on the intercepted tape, or even created a reasonable doubt as to that question, his client would have been acquitted. Even were this fact, standing alone, not sufficient to compel the average competent attorney in this jurisdiction to conduct' a thorough aural comparison of the voice exemplar with the intercepted recording, we have no doubt that such a duty arose after Trice repeatedly asserted (even to the point of rejecting a plea agreement) that his was not the voice on the intercepted conversation, and especially after one of Trice’s co-defendants informed trial counsel that he (the co-defendant) and not Trice was the speaker in the phone call under consideration. The mere possibility that investigation of the exemplar might have produced nothing of consequence for Trice’s defense cannot serve as a justification for the failure to perform such an investigation in the first place.9
At bottom, our conclusion in this regard is grounded on a standard which is central to the successful functioning of our adversary system of criminal justice: a defense attorney, whether appointed or retained, is obligated to inquire thoroughly into all potentially exculpatory defenses and evidence. “It is the duty of the lawyer to conduct a prompt investigation of the circumstances of the case and explore all avenues leading to facts relevant to guilt and degree of guilt or penalty.” American Bar Association Standards, supra, at § 4.1 (emphasis added). This investigation, among other things, “should always include efforts to secure information in the possession of the prosecution and law enforcement authorities.” Id. (emphasis added). If our society is to retain its commitment to the not-unattainable ideal that each accused, however destitute, ignorant, or seemingly culpable, is entitled to a complete and vigorous defense, the courts must continue to insist that trial counsel, at the very least, investigate all substantial defenses available to a defendant.10
In sum, we believe that under the circumstances of this case, an attorney exercising customary skill and knowledge would have aurally11 compared the inter[669]*669cepted recording with the voice exemplar in the hope of obtaining exculpatory evidence to present to the jury. Accordingly, we hold that trial counsel’s acknowledged failure so to proceed on behalf of Trice constitutes ineffective assistance of counsel.
Ill
Having determined that Trice was ineffectively represented at trial, we proceed to consider whether the defendant suffered prejudice as the result of his attorney’s constitutional negligence. Initially, however, we confront Trice’s claim that given a finding of ineffective assistance, a showing of prejudice is unnecessary.
A
Trice contends that once the district court concluded that his representation had been ineffective, habeas relief should have been granted as of course. He argues that recent Supreme Court cases indicate that a showing of prejudice is not requisite with respect to ineffective assistance matters. In particular, Trice relies on Geders v. United States, 425 U.S. 80, 96 S.Ct. 1330, 47 L.Ed.2d 592 (1976); Herring v. New York, 422 U.S. 853, 95 S.Ct. 2550, 45 L.Ed.2d 593 (1975); and Holloway v. Arkansas, 435 U.S. 475, 98 S.Ct. 1173, 55 L.Ed.2d 426 (1978), which, in his view, stand for the no-harmless-error proposition.
The cases cited by Trice, however, are concerned not with the ineffective assistance of counsel, but rather, in various settings, with the denial of counsel altogether. In Geders, counsel was prohibited by the trial judge from communicating with his client during an overnight recess; the Court held that for this period, the defendant’s “ ‘right to be heard by counsel’ ” had been abridged. 425 U.S. at 88-89, 96 S.Ct. at 1335-1336 (quoting Powell v. Alabama, 287 U.S. 45, 68, 53 S.Ct. 55, 63, 77 L.Ed. 158 (1932)). Similarly, in Herring, the Justices concluded that a defendant had been denied his right to the “Assistance of Counsel” where the court, acting pursuant to a state statute, barred an attorney from presenting a closing argument. 422 U.S. at 857, 95 S.Ct. at 2552. And in Holloway, the Court held that a defendant’s right to counsel had been infringed where his attorney improperly also represented a codefendant; in such a situation, prejudice “is presumed regardless of whether it was independently shown,” because a defendant represented by a lawyer serving two clients with contradictory interests in effect is denied his right to counsel altogether. 435 U.S. at 489, 98 S.Ct. at 1181. See generally Decoster, supra, 624 F.2d at 200-02; Cooper v. Fitzharris, 586 F.2d 1325, 1332 (9th Cir. 1978) (in banc), cert. denied, 440 U.S. 974, 99 S.Ct. 1542, 59 L.Ed.2d 793 (1979). These decisions, we believe, are offspring of the general rule, first announced in Gideon v. Wainwright, 372 U.S. 335, 83 S.Ct. 792, 9 L.Ed.2d 799 (1963), that the complete denial of counsel to a defendant could never give rise to harmless error. See Chapman v. California, 386 U.S. 18, 23 n.8, 87 S.Ct. 824, 827 n.8, 17 L.Ed.2d 705 (1967); see also United States v. Welty, 674 F.2d 185, 194 n.6 (3d Cir. 1982); United States v. Laura, 607 F.2d 52, 58 (3d Cir. 1979).
Although no decision of the Supreme Court has dealt squarely with the showing required of a petitioner making the usual ineffective assistance claim under the habeas statute,12 arguably that question was resolved in United States v. Morrison, 449 U.S. 361, 101 S.Ct. 665, 66 L.Ed.2d 564 (1981). In Morrison, the Supreme Court, reversing this Court, 602 F.2d 529 (1979), held that an indictment could not be dismissed for government misconduct absent an allegation and showing “that the claimed violation had prejudiced the quality or effectiveness of respondent’s legal representation.” Id. at 363, 101 S.Ct. at 667. In arriving at this result, the Justices observed that
[670]*670[c]ases involving Sixth Amendment deprivations are subject to the general rule that remedies should be tailored to the injury suffered from the constitutional violation and should not unnecessarily infringe on competing interests.... Our approach has thus been to identify and then neutralize the taint by tailoring relief appropriate in the circumstances to assure the defendant the effective assistance of counsel and a fair trial. The premise of our prior cases is that the constitutional infringement identified has had or threatens some adverse effect upon the effectiveness of counsel’s representation or has produced some other prejudice to the defense.
Id. 449 U.S. at 364-65, 101 S.Ct. at 667-68 (emphasis added). See also Chambers v. Maroney, 399 U.S. 42, 90 S.Ct. 1975, 26 L.Ed.2d 419 (1970), where the Court declined to “fashion a per se rule requiring reversal of every conviction following tardy appointment of counsel or to hold that, whenever a habeas corpus petition alleges a belated appointment, an evidentiary hearing must be held to determine whether the defendant has been denied his constitutional right to counsel.” Id. at 54, 90 S.Ct. at 1982.
Even if the Supreme Court’s opinions in Morrison and Chambers v. Maroney were to be limited to their particular facts, however, the harmless error standard of Chapman v. California, supra, applied by the district court in the proceeding below, would appear to govern. Under Chapman, “before a federal constitutional error can be held harmless, the court must be able to declare a belief that it was harmless beyond a reasonable doubt.” 386 U.S. at 24, 87 S.Ct. at 828. Rather than granting collateral relief outright when a conviction is tainted by ineffective assistance, then, a court should expect the habeas petitioner to demonstrate that there is a “reasonable possibility” that had the error of which he complains not occurred, the jury might have arrived at a different outcome.
Consistent with this analysis, our Court has long required a habeas corpus petitioner to demonstrate prejudice in ineffective assistance situations.13 See Boyer v. Patton, 579 F.2d 284, 288-89 (3d Cir.), cert. denied, 425 U.S. 997, (1976); United States ex rel. Johnson v. Johnson, 531 F.2d 169, 177-78 (3d Cir.), cert. denied, 425 U.S. 995, (1976); United States v. Crowley, 529 F.2d 1066, 1070-71 (3d Cir. 1976); United States ex rel. Green v. Rundle, 434 F.2d 1112, 1115 (3d Cir. 1970); accord, Decoster, supra, 624 F.2d at 202-09; Cooper v. Fitzharris, supra, 586 F.2d at 1331-33; McQueen v. Swenson, 498 F.2d 207, 218-20 (8th Cir. 1974). In fact, in our previous opinion in this very case, we ordered that a hearing be conducted to determine “whether counsel’s failure to explore the exemplar issue prejudiced Trice’s case.” Baynes, 622 F.2d at 70 (emphasis added). We cannot agree, therefore, with Trice’s contention that the district court erred in insisting upon a showing of prejudice in this case.14
As we recognized in United States ex rel. Green v. Rundle, supra, however, “[i]n many instances ineffective assistance of counsel may have had so pervasive an effect on the process of guilt determination that it is impossible to determine accurately the presence or absence of prejudice.” 434 F.2d at 1115. In such cases, any substantial doubt must be resolved in favor of the defendant, inasmuch as Chapman permits a finding of harmless error only where it is concluded beyond a reasonable doubt that no prejudice resulted from the identified [671]*671constitutional violation.15 To prevail on this appeal, Trice need not prove that it was not his voice on the intercepted recording; instead, he need only show that his trial attorney’s “exploration of the voice exemplar issue might have led to a viable defense and a verdict favorable to [him],” Baynes at 69 (emphasis added). In other words, Trice need only demonstrate that his attorney’s ineffectiveness was not harmless “beyond a reasonable doubt.”
B
In determining that Trice was not prejudiced by his attorney’s failure to compare the exemplar with the intercepted recording, the district court relied on two factors: a spectrographic analysis, which concluded that the two tapes reflected “one and the same” speaker, and an aural comparison, which produced, in the judge’s view, only “de minimis” pronunciation variations.16 After evaluating these factors, the court “found” that the jury’s decision to convict Trice would not have been altered even had defense counsel investigated and exploited the exemplar evidence. We now examine both of the factors that were deemed dispositive by the district court.
The spectrographic analysis. The district court concluded that Trice was unable to establish prejudice because “[t]he results of the spectrograph analysis suggest that [he] was a speaker in the subject incriminating telephone conversation and, therefore, a scientific comparison by trial counsel would not have yielded evidence of an exculpatory nature or of a nature possibly altering the jury’s decision.” For a number of reasons, we believe this conclusion is logically insupportable.
First,' nothing in the record indicates that a spectrographic analysis is “foolproof.” In fact, the reliability of the procedure is the subject of considerable dispute. As the district court itself recognized, “it is far from resolved whether [spectrographic] evidence [is] reliable to the extent justifying its use.” And, as the Government notes, the admissibility of such evidence is a question that has not been resolved in this Circuit.17 We need not decide today whether spectrographic analyses are reliable or whether spectrograms are admissible in criminal trials. For present purposes, it is sufficient to observe that the spectrographic process is [672]*672not claimed to be infallible, and those courts which have permitted introduction of spectrograms have stressed that their reliability and significance are matters for resolution by a jury. Under these circumstances, we believe that the district court erred in attaching controlling weight to the spectrogram prepared in this case. Although there is nothing in the record to suggest as much, we simply cannot rule out the possibility that the spectrographic assessment prepared here is unreliable by its very nature. This being the case, it is not possible to conclude, on the basis of one spectrographic analysis, that Trice suffered no prejudice as a result of his lawyer’s ineffectiveness.
Second, and as a related matter, the district court erred in accepting the results of the spectrographic analysis without examining how the expert arrived at his conclusion, and without providing the defense with an opportunity to contest that expert’s qualifications, procedures, or findings. Again, although nothing in the record so indicates, we cannot dismiss the possibility that the spectrogram produced for this case was prepared improperly or carelessly. Cross-examination of the spectrographic expert before a jury might reveal defects in his procedure or analysis — defects that might serve to create a “reasonable doubt” whether the voice exemplar constituted exculpatory evidence.
Finally, even were we to conclude' that spectrographic analyses in general — or this spectrogram in particular — are reliable sources of evidence, we agree with Trice that the emphasis on the results of the spectrographic assessment tended to obfuscate the ineffective assistance issue. As we have observed, see note 11 supra, trial counsel was under no obligation, in 1975, to procure a scientific comparison of the two recordings before proceeding to trial. In the absence of such a duty, the results of a spectrographic analysis, performed a half dozen years later, would not seem relevant in determining ineffectiveness at the time of trial. If indeed it is established that Trice suffered harm because his lawyer did not make an aural comparison of the two tapes, we do not believe that such a showing of prejudice can be overcome by focusing on the product of subsequent scientific advances utilized six years after the trial was completed.
The aural comparison. The district court also “found” that Trice was not prejudiced by his attorney’s shortcomings because a jury would have dismissed the pronunciation differences contained in the two tapes as “de minimis.” We agree with Trice, however, that in making this determination, the district court in effect was invading the province of the jury. Unlike the district judge, we simply are not prepared to hold, as a matter of law, that given the present record, a jury could not entertain “reasonable doubts” whether Trice’s voice was on the intercepted tape. In arriving at this result, we are influenced by four facts: a) three words were pronounced divergently on the two recordings; b) the Government’s witness testified that it was “absolutely” possible for someone to conclude that the two tapes contained different voices, and that he would “lend that opinion due respect”; c) a co-defendant of Trice admitted to trial counsel that it was his, and not Trice’s voice on the intercepted conversation; and d) Trice apparently never took the prison visit to which the telephone conversation referred. It is quite conceivable that a jury, after listening to the two tapes, and after hearing counsel stress these four facts, might acquit Trice.18 It is also conceivable, of course, that a jury might dismiss the pronunciation differences as “de minimis,” conclude that Trice intentionally distorted his voice when recording [673]*673the exemplar, believe that the co-defendant was lying in order to protect Trice, and reject the prison visitation evidence as irrelevant. The point, however, is that precisely questions of this sort — the resolution of which ultimately will affect judgments relating to innocence and guilt — are to be weighed and evaluated by a jury after a trial, and not by a court during a habeas proceeding.
Given that the intercepted tape was the only evidence offered against Trice during the course of a nineteen day trial, it is not difficult to envision how an effective trial counsel might have exploited the exemplar evidence to great effect in order to advance his client’s cause. While we cannot say with any certainty that, as the result of such a performance, Trice would have been acquitted, that is not the test. All we need determine is that Trice’s defense was prejudiced by his counsel’s ineffectiveness — i.e., that the exemplar evidence, if investigated, “might have led to a viable defense and a [favorable] verdict,” Baynes, 622 F.2d at 69 (emphasis added), and that the failure of Trice’s trial attorney so to proceed is not harmless “beyond a reasonable doubt,” Chapman, 386 U.S. at 24, 87 S.Ct. at 828. We so conclude.
IV
The judgment of the district court will be reversed and the cause remanded with instructions to enter an order granting the writ of habeas corpus and directing that a new trial be held within a reasonable time.