I am unable to agree with the court that admitting testimony of the scout car identification was not constitutional error.
I
The heart of the court’s decision lies in its conclusion that Officer Mitchell’s conduct in this case was reasonable:
He said that he knew nothing about the crime for which appellant was wanted. Accordingly, it was not, nor could it have been, his purpose to link appellant to it. His immediate job, as he conceived it, was simply to be sure that he had picked up the man who was the subject of the radio lookout.1
No claim seems to be made here that Officer Mitchell made an honest mistake and thought that the situation was no different from Russell; the court does not argue that he believed he was returning appellant to the scene of the crime minutes after its commission.2 *170We are faced in this ease, therefore, with a police officer who knew that he was not within the Russell rule, but who nevertheless returned the suspect to the complaining witness for a one-man show-up. Given this context, I think it is wrong to identify the reasonableness of the police officer’s conduct as the essential issue in this case.
It was not absolutely clear to Officer Mitchell that he had in fact arrested the right man — that is, the man whom the complaining witness had spotted a few minutes before. Hence his desire to return to the scene of the identification for immediate confirmation. Whether appellant was the man complainant had just seen on the street would later become a jury question, a necessary step towards the conclusion that appellant was the man who assaulted complainant two weeks before. Viewed in this light, the situation does look very much like Russell: the suspect was not being returned to the scene of the crime immediately after its commission, but he was being returned to the scene of an on-the-street identification, immediately after it occurred. The question, then, is whether we carve out an exception to the Wade rule parallel to the Russell exception. To say that the police conduct here was reasonable does not answer this question; it either begs the question or states only part of it. Our decision here should turn on a recognition of the rules laid down by the Supreme Court in Wade, on an examination of the principles and policies that underlie those rules, and on a judgment whether those princples and policies lack sufficient application to the situation at hand so that there is no point in mechanically applying the rules here. The court proceeded in that fashion in Russell, and it should do so in this case. To me, application of that method leads to the conclusion that we must find constitutional error.
II
In Russell v. United States 3 we held that Wade 4 does not require the exclusion of prompt on-the-scene identifications of a freshly apprehended suspect. This exception was later extended to cover similar identifications where the witness is taken to the scene of the arrest.5 We recognized that permitting such procedures seemed to fly in the face of the Supreme Court’s manifest concern with any pretrial confrontation.6
The Court had said:
[T]he confrontation compelled by the State between the accused and the victim or witnesses to a crime to elicit identification evidence is peculiarly riddled with innumerable dangers and variable factors which might serious*171ly, even crucially, derogate from a fair trial.7
These dangers, needless to say, are particularly acute when a single suspect is viewed in police custody. In Russell, we found a balance to these dangers in the greater accuracy of fresh identifications made within minutes of the witnessed crime.
A man may see clearly in his “mind’s eye” a face or a figure which he is hard put to describe adequately in words. Though the image of an “unforgettable face” may occasionally linger without any translation into words, photographic recall is most often ephemeral. Vivid in the flash of direct observation it fades rapidly with time. And the conscious attempt to separate the ensemble impression into particular verbalized features, in order to preserve some recollection, may well distort the original accurate image so that it is the verbalized characteristics which are remembered and not the face or the man.
Balancing all the doubts left by the mysteries of human perception and recognition, it appears that prompt confrontations in circumstances like those of this case will “if anything promote fairness, by assuring reliability * * *.”8
Clearly this justification for an exception to Wade is absent in this case. Two weeks had passed since the crime. This is ample time for a witness’s mental image of the criminal to fade or to shift imperceptibly as it is reflected upon. And while a spontaneous identification at a later date may be evidence that a vivid recollection has been preserved, there is not the guarantee obviously present in Russell that whatever vivid image is preserved in the mind during the “few moments” between sighting and arrest is in fact the image of the criminal.
Even though the specific justification for the Russell exception may not apply to this case, it may seem excessively technical to apply Wade to a scout car identification which follows a spontaneous identification on the street; for the spontaneous identification alone may “taint” the witness’s later line-up or in-court identification. For example, the witness may identify the defendant on the street and, by getting a close look at him, observe particular features or characteristics that she had not noticed when the crime was committed. She may then rely upon these features when she picks the defendant out of a line-up or when she identifies the defendant in court. In such a case, an intervening scout car identification — however suggestive — would not be contributing to the later identification.
We cannot say, however, that spontaneous identifications will generally taint later identifications in the manner described and that the scout car identification will have no effect. In the first place, the witness may not have got a close look at the man she picked out on the street. If he was across the street, as he was in this ease, or if his features were partially concealed by a hat, the witness may have decided that it is highly likely that he is her assailant, but still want to get a closer look before she is absolutely certain. It is precisely this “closer look” which may be prejudicial, given the suggestive nature of scout car identifications. Even where the witness thinks she has been positive in her identification on the street, a scout car confrontation may allay doubts of which she is not fully conscious. Where the witness has called the police to come and arrest the man she suspects, fear of embarrassment alone may subtly encourage her to identify the man they apprehend and then remember him in subsequent identifications.
In sum, scout car confrontations after spontaneous identifications on the street *172do not generally assure reliability to such a degree that another exception to the Wade rule is justified.
It is tempting, at this point, to be swayed by the argument that prompt return to the scene in cases like this one will reduce unnecessary detentions of innocent suspects. As an initial matter, it must be emphasized that in Russell, though “expeditious release of innocent suspects” was mentioned as a factor in the decision,9 it was considered as a factor only after it had been determined that the circumstances of the confrontations in question “if anything promote fairness, by assuring reliability.” 10 For the reasons I have already given, I cannot say that circumstances surrounding confrontations similar to the one in this case will assure reliability, nor do I find any such conclusion in the opinion of the court.11 Russell, therefore, lends no support to the court’s invocation of “protecting the innocent.”
Even apart from the applicability of Russell, however, the argument from reducing unnecessary detention of innocent suspects is very weak. First, I believe that an innocent man, were he as aware of the dangers of misidentification as lawyers and judges are, would generally prefer to have the necessary time taken for a fairly conducted lineup. The logic of protecting the innocent forces one to the position that the supposedly innocent man should have some voice in the decision to return to the complaining witness. “Protecting the innocent” by giving them no choice at all has a very hollow ring to me.
Second, it seems to me that innocent men are adequately protected by the requirement that the police have probable cause before they arrest a suspect. In this case, for example, appellant was arrested within minutes of the' time he had been sighted and described to the police as the perpetrator of a violent crime; he was spotted by an officer and arrested within two blocks of that initial sighting, wearing clothes which precisely fit the description given by complainant; he was first spotted in the direction in which complainant and her companion said he had fled, and his actions upon seeing the police car were suspicious. There is not the slightest doubt that there was probable cause to arrest this suspect, and therefore, probable cause for holding him a reasonable length of time to conduct a line-up.12 In fact, the presentation of the innocent suspect argument in this particular case shows how easy it is to neglect and unnecessarily compromise the Wade principle. Not only was there ample probable cause to arrest this suspect, but there is no indication that other men seen in the area also fit the description given by the complainant.13 Furthermore, the time *173was only 5:40 in the afternoon, hardly an hour at which the difficulty of preparing a prompt line-up, if the police had substantial doubts about the guilt of the man they have just arrested, seems extraordinary.14
I conclude that the scout car identification in this case did not occur in the kind of situation for which “substantial countervailing policy considerations” permit another general exception to the Wade requirement of counsel.15 I would hold that it was constitutional error to allow the Government to introduce evidence of that identification into its case.
Ill
Finding a Wade violation does not automatically require reversal of the conviction. It remains for me to state why I do not think that admission of the improper evidence in this case was harmless error. Given the strict standards laid down in Chapman v. California, 386 U.S, 18, 87 S.Ct. 824, 17 L.Ed.2d 705 (1967), and Harrington v. California, 395 U.S. 250, 89 S.Ct. 1726, 23 L.Ed.2d 284 (1969), it will be exceedingly rare to find harmless error when that error concerns the identification testimony of the sole witness and victim of a crime, and where no other evidence links appellant to that crime. In such circumstances, a claim of harmless error would not ordinarily deserve serious consideration; but the peculiar facts of this case do force me to examine closely the possibility of harmless error.
The critical question, as I understand Chapman and Harrington, is not whether it is clear beyond a reasonable doubt that the jury would have convicted appellant if it had not heard the illegally admitted evidence, even though members of this court — the author of this opinion included — have occasionally spoken as if it were.16 Instead, I think it is clear that the test must be whether the appellate court is convinced beyond a reasonable doubt that the illegally admitted evidence did not contribute to the verdict, or alternatively, did not sway some members of the jury17 In short, what *174is critical is the impact of the illegal evidence on the jury, not the weight of the untainted evidence.
I draw these conclusions from essentially no more than a careful reading of Chapman and Harrington. These opinions are brief, so there is no point in quoting extensively from them here. It is clear that the Court in Chapman explicitly turned away from what may be called the “overwhelming evidence test” and associated itself with the test set out in Fahy v. Connecticut, 375 U.S. 85, 84 S.Ct. 229, 11 L.Ed.2d 171 (1963): “whether there is a reasonable possibility that the evidence complained of might have contributed to the conviction.”18 Mr. Justice Brennan, dissenting in Harrington, accused the majority of overruling Chapman and adopting the overwhelming evidence test; but whatever ambiguity exists in the majority opinion does not permit this court to agree with him. The Court said at the end of its opinion: “We do not depart from Chapman-, nor do we dilute it by inference. We reaffirm it.”
The difference between the two tests is small, and the scopes of inquiry overlap. The weight of the untainted evidence in any case is certainly relevant to determining the possible impact on the jury of certain illegally admitted evidence. Nevertheless, there will be cases where the other evidence of guilt seems overwhelming, but where the illegal evidence was of a particularly persuasive kind. Harrington and Chapman require that these be reversed. We might recall the words of Justice Rutledge, discussing the federal harmless error rule in Kotteakos v. United States, 328 U.S. 750, 763-764, 66 S.Ct. 1239, 1247, 90 L. Ed. 1557 (1946):
[I]t is not the appellate court’s function to determine guilt or innocence. * * * Nor is it to speculate upon probable reconviction and decide according to how the speculation comes out. Appellate judges cannot escape such impressions. But they may not make them sole criteria for reversal or affirmance. Those judgments are exclusively for the jury, given always the necessary minimum evidence legally sufficient to sustain the conviction unaffected by the error. * * *
Turning now to the case at hand, I think, initially, that in the absence of testimony about the scout car identification, some jurors would have felt compelled to look much more closely at the events immediately following complainant’s sighting the man in the alley across the street. A juror might not seriously doubt the correctness of that initial identification but still wonder whether the police arrested the right man. What might stand out in his mind is that appellant, when first seen by Officer Mitchell, seems to have been walking along the street in the opposite direction from that in which one would expect a man to be going if he were fleeing from the place where complainant first saw him, in the direction in which complainant claims to have seen him run. Not only that, but he was still only two or three blocks (depending upon how one postulates his path) from the place of sighting. Since both complainant and her companion agree that the man ran off, the time between her sighting and the initial police dispatch becomes important for a determination whether it is likely that a fleeing man would still be so close. An implicit contradiction that seems apparent between the testimony of complainant and her companion as to what street the fleeing man turned down only complicates the matter further. I recognize that according to Officer Mitchell appellant ran when he saw the police car and that appellant’s own explanation of his actions at that time is implausible. Nevertheless, given the puzzling fea*175tures of the chase, I feel that there is a very real possibility that what swayed some members of the jury on this issue was complainant’s prompt identification right after appellant was arrested.
In addition, I cannot rule out the possibility that some members of the jury had qualms about the spontaneous identification made late in the afternoon from across the street. These jurors, too, might have had their doubts allayed by the scout car identification.
Since I cannot say beyond a reasonable doubt that the error in admitting testimony about the scout car identification did not contribute to the verdict, I would reverse these convictions and remand for a new trial.
. P. 167 supra.
. The record does not justify a conclusion that Officer Mitchell thought he was in a Russell situation, and if the court had wished to rely on such a possibility, it would have had to remand the case for further findings.
The Government had clearly had hopes of suggesting that such an honest mistake had been made by Officer Mitchell, as its line of questioning at the identification hearing shows:
Q Now, there was a radio run that said he was being chased by someone from the scene of an assault, is that correct, sir?
A [Officer Mitchell] It didn’t say he was being chased from the scene of an assault; he was being chased by a subject and he was wanted for assault.
It is also worth remarking that the initial police dispatch, p. 164 n. 1 supra, stated (inaccurately) that the offense took place “last week,” and Officer Mitchell, who *170clearly heard an address (which he asked to have retreated) and who heard and remembered the description of the suspect, never said or even suggested that, he failed to hear that particular part of the dispatch.
. 133 U.S.App.D.C. 77, 408 F.2d 1280. cert. denied, 395 U.S. 928, 89 S.Ct. 1786, 23 L.Ed.2d 245 (1969).
. United States v. Wade, 388 U.S. 218, 87 S.Ct. 1926, 18 L.Ed.2d 1149 (1967).
. United States v. Miller, No. 22,332 (D. C.Cir., March 23, 1970).
. “[T]he principle of Powell v. Alabama, 287 U.S. 45, 53 S.Ct. 55, 77 L.Ed. 158, and succeeding cases requires that we scrutinize any pretrial confrontation of the accused to determine whether the presence of his counsel is necessary to preserve the defendant’s basic right to a fair trial as affected by his right meaningfully to cross-examine the witnesses against him and to have effective assistance of counsel at the trial itself. It calls upon us to analyze whether potential substantial prejudice to defendant’s rights inheres in the particular confrontation and the ability of counsel to help avoid that prejudice.”
United States v. Wade, 388 U.S. at 227, 87 S.Ct. at 1932 [emphasis in original].
. Id. at 228, 87 S.Ct. at 1933, quoted in Russell v. United States, supra, 133 U.S.App.D.C. at 79, 408 F.2d at 1282.
. 133 U.S.App.D.C. at 81, 408 F.2d at 1284 [emphasis added; footnote omitted].
. Id.
. See p. 171, supra.
. On page 168, the court does ask whether Officer Mitchell’s action gave rise “to a very substantial likelihood of irreparable misidentification.” To be sure, if this question were answered in the affirmative, his conduct is not likely to seem reasonable. Quite correctly, the court answers the question in the negative. The difficulty, however, is that this question is relevant to a challenge to the identification as a violation of Stovall v. Denno, 388 U.S. 293, 87 S.Ct. 1967, 18 L.Ed.2d 1199 (1967), but it is not relevant to a challenge based upon Wade, where the proper question is whether the circumstances were so conducive to reliability of identification that there is no reason to apply the Wade rule.
. It might be noted that lack of an adequate description to furnish the police with probable cause to arrest need not hamstring them entirely. In Stewart v. United States, 135 U.S.App.D.C. 274, 418 F.2d 1110 (1969), police officers took the complainant with them in the police ear and cruised the neighborhood for an hour and a half before spotting three men who fit complainant’s descriptions. If complainant had not been able to furnish a verbal description, it may still have been permissible to take him in the police car for a search of the neighborhood.
. The absence of other men in the area fitting the description further supports a finding of probable cause for the arrest of appellant. But it is perfectly possible that a complainant’s description might *173be comparatively detailed, and that two men found in the immediate vicinity fit the description. In such a case there may well be probable cause to arrest both men and put them in a line-up, even though it is clear that one of them must be innocent.
. Cf. Stewart v. United States, 135 U.S.App.D.C. 274, 418 F.2d 1110 (1969) (confrontation at 4:00 a. m.) ; Russell v. United States, supra (5:00 a. m.).
. See Russell v. United States, supra, 133 U.S.App.D.C. at 80, 408 F.2d at 1284, quoting United States v. Wade, 388 U.S. at 237, 87 S.Ct. 1926.
Nor do I think it appropriate to affirm this case as a particular exception to the Wade rule. Spontaneous identifications by victims a substantial length of time after the crimes have been committed are not extraordinary. See e. g., United States v. Green, 141 U.S.App.D.C. 136, 436 F.2d 290 (1970) discussed in the court’s opinion, supra. More important, I fear that ad hoe determinations of exceptions to the per se rule in Wade would soon tend to undercut it substantially.
. E. g., United States v. Miller, No. 22,-332 (March 23, 1970) (Bazelon, C. J.) slip op. at 9: “We cannot imagine a different result if Goldberg’s testimony had been barred”; Taylor v. United States, 134 U.S.App.D.C. 246, 248-249, 414 F.2d 1142, 1144-1145 (1969) ; “In applying Ghapman we must look to all the evidence, defense and prosecution alike, and bring our judgment to bear upon the question of whether it is clear to us beyond a reasonable doubt that a guilty verdict would have resulted even if the jury had never heard the challenged testimony.” I make no suggestion at all that these decisions were incorrect, or even that an improper test was in fact applied.
. See, e. g., Hamilton v. United States, 139 U.S.App.D.C. 368, 433 F.2d 526 (D.C. Cir., July 31, 1970) at 531: “[W]e find beyond a reasonable doubt that, because of the overwhelming weight of the evidence, the admission of Klein’s out-of-court statements did not affect the jury’s verdict * *
. 375 U.S. at 86-87, 84 S.Ct. at 230.