concurring in part and dissenting in part:
I concur in the majority’s holding regarding Rule 33 but respectfully dissent from its ruling under Brady v. Maryland, 373 U.S. 83, 83 S.Ct. 1194, 10 L.Ed.2d 215 (1963).
I
Jernigan was initially charged with three robberies — a September 20, 2000 robbery at 15 East Guadalupe in Gilbert; an October 11, 2000. robbery at 906 East Baseline Road in Tempe; and an October 25, 2000 robbery at 2298 North Alma School in Chandler.1 After Jernigan was placed into custody, two more banks were robbed in close proximity to the 15 East Guadalupe branch, by a person matching Jernigan’s physical description — a short, Hispanic woman with acne. Those robberies took place on November 28, 2000 and November 30, 2000.
I find it highly unlikely that in a case like this, which turns 'exclusively and entirely upon eyewitness testimony, the jury would have convicted Jernigan had it known that robberies at nearby banks by a short Hispanic woman with acne continued after Jernigan’s arrest. But the prosecution never disclosed this information to Jernigan’s defense team and never bothered to investigate whether this second robber — not Jernigan — actually committed the crime for which Jernigan was charged. The government’s failure to disclose this highly relevant information about the subsequent robberies violates its obligations under Brady, causing me to respectfully dissent.
II
A Brady violation takes place when the government withholds material, exeulpato-ry evidence. “To establish a Brady violation, the evidence must be (1) favorable to the accused because it is either exculpatory or impeachment material; (2) suppressed by the government, either willfully or inadvertently; and (3) material or prejudicial.” United States v. Blanco, 392 F.3d 382, 387 (9th Cir.2004).
Information regarding the additional robberies was clearly favorable to Jerni-gan’s case and never disclosed to the defense; consequently, Brady’s first two elements are established. The lapse was also material: the similarities in identity are uncanny, and the prosecution relied exclusively on visual-identification evidence. Thus, the possibility of wrongful conviction always lurked in the background of this case, and, plainly, there was “ ‘reasonable probability’ that had the evidence been disclosed the result at trial would have been different.” Wood v. Bartholomew, 516 U.S. 1, 5, 116 S.Ct. 7, 133 L.Ed.2d 1 (1995).
The majority takes comfort in the testimony of a handful of eyewitnesses, but I am not so easily persuaded. “Centuries of experience in the administration of criminal justice have shown that convictions based solely on testimony that identifies a defendant previously unknown to the witness is highly suspect. Of all the various kinds of evidence it is the least reliable, especially where unsupported by corroborating evidence.” Jackson v. Fogg, 589 F.2d 108, 112 (2d Cir.1978). Regrettably, the majority ignores this vital admonition.
Ill
Jernigan was questioned because she was suspected of shoplifting and was a small Hispanic woman, not because of any *1035known connection to bank robberies. The Government presented no evidence of any kind to corroborate visual identifications of Jernigan save an F.B.I. analysis of a surveillance video pegging the robber’s height at just under five feet. The fingerprint lift did not match Jernigan’s print, and the police found no firearms, clothing, money, or other items associating Jernigan with the robbery.
Subsequent to Jernigan’s trial, Rodriguez-Gallegos was charged with the November 28 and 30 robberies, as well as a December 11, 2001 robbery of the same bank that Jernigan had allegedly robbed. By the time of the arrest, Jernigan had been convicted. Still, the prosecution could have — but did not — conduct any inquiry into whether Rodriguez-Gallegos had committed the robbery for which Jer-nigan was charged. It could have at least checked the fingerprint lift from the September 20 robbery with Rodriguez-Gallegos’s print. Such tunnel vision puts into question the fair administration of justice.
Equally troubling are the inconsistencies within the eyewitness accounts. Elizabeth Chlupsa, the victim-teller on September 20, described the perpetrator as having plucked eyebrows, wearing “very very very dark eyeliner,” “a lot of make-up,” and a lot of eyeliner. By contrast, Lorraine Hawley, who said she looked at the perpetrator for an extended period of time, described the robber as wearing “little to no make-up” and no lipstick.2
Chlupsa indicated both immediately after the robbery and during trial that the perpetrator had no tattoos and did not have painted fingernails. However, Jerni-gan has a number of tattoos on her hands and forearms, as evidenced in photographs submitted to the district court and confirmed by the FBI shortly after the arrest.
The eyewitnesses’ tentativeness is equally telling. Hawley, who claimed to get a good look at the robber, had some trouble picking Jernigan out of a photospread, stating, “I want to say it looks like this one.” And Golliher appears to have had trouble as well, stating, “By — I just felt that was the one that — that was whom I saw.”
These problems are not altogether surprising given that all but one of the eyewitnesses viewed the photospread five to six months after the incident. Lorraine Haw-ley saw the photospread two days before trial — six months after the robbery. Three others — Kathleen Golliher, employee Yarjanic Nath, and customer Donovan Grierson — were all shown the photos five months after the robbery. This delay, which goes totally unexplained, also detracts from the reliability of the witness identifications.
Additionally, the district court excluded evidence conducted by a polygraph expert that supports Jernigan’s claim that she was not the September 20 robber. The polygrapher asked Jernigan about the robbing of banks both as a general matter and in regard to the three banks for which she was originally arrested. The polygrapher listed each one of the banks, and Jernigan denied any involvement in those or other bank robberies. Her score of 17 indicates a truthful response (anything above a score of six is considered truthful). However, the district court excluded documentary evidence of the polygraph tests and precluded the polygrapher from testifying.
*1036Finally, one cannot avoid the extremely low likelihood, based on crime statistics, that two short female Hispanic bank-robbers with acne would even exist in such a small geographical area. In 2000 (the year all robberies in question took place), only six percent of all bank-robbery perpetrators were female; moreover, only six percent of bank-robbers overall (male and female) were Hispanic. See Summary and Interpretation of Bank Crime Statistics, WOO (Federal Bureau of Investigation, May 12, 2002). The likelihood of two small Hispanic female robbers holding up banks, let alone the same bank and other banks in the same area, is therefore extremely low.
IV
If these problems with the trial — which the parties do not dispute — are not enough to warrant reversal, I’ll pass briefly over additional disputed items that give me further reason for pause. Jernigan alleges that certain witnesses were not properly advised that the person who had robbed the bank might not be depicted in one of the six photographs; that certain eyewitnesses used a “process of elimination” approach in selecting Jernigan from the pho-tospread; and that the modus operandi of all of the 2000 robberies bears a striking resemblance, as all involved a hand-written demand note, on folded paper, warning the teller not to set off any alarms.3
V
When I consider (1) the lack of any evidence to corroborate visual identifications (save an F.B.I. analysis of a surveillance video pegging the robber’s height at just under five feet), (2) the lack of any firearms, clothing, money, or other items associating Jernigan with the robbery, and (3) most importantly, the existence of another person matching Jernigan’s physical description — a short, Hispanic woman with a pock-marked face — robbing nearby banks during the same time-period, in a case based entirely on visual identifications, I cannot have any confidence in the eyewitness identifications.
In light of all the above, we should simply remand for a new trial. If a jury still finds the prosecution’s case compelling, that ends the matter, and the prosecution and this court can feel assured that the right person(s) remain(s) incarcerated. If it acquits, justice will have been served. If we don’t remand, we must live with the doubts sown by the Brady violations and the anomalies in this unusual case, despite the nagging suspicion that mistakes might have been made.
VI
Although I’m very concerned that an innocent person may have been wrongly convicted, that is not the critical point driving my dissent. Under Brady, the government had a duty to advise the defense of the continuing robberies. That information very likely could have affected the verdict. Jernigan was clearly deprived of “favorable evidence [that] could reasonably be taken to put the whole case in such a different light as to undermine confidence in the verdict.” Kyles v. Whitley, 514 U.S. *1037419, 435, 115 S.Ct. 1555, 131 L.Ed.2d 490 (1995) (footnote omitted).
Further, of course, the suppression undermines confidence in the outcome of the trial and the integrity of the government as well as its prosecutors’ dedication to justice. A new trial is clearly warranted. “[I]t is a miscarriage of justice ... to convict an innocent person.” United States v. Ameline, 409 F.3d 1073, 1074, 1081 (9th Cir.2005) (en banc) (citation omitted). In light of the uncertainties as to identity, I would order a new trial for Jernigan. I therefore respectfully dissent.
. Jernigan was tried only for the firsj robbery (the September 20 robbery) and an associated 924(c) count.
. Witnesses seem to have had trouble determining whether the robber was Hispanic or Asian. Although I place less emphasis on these inconsistencies in identification, they, too, deserve mention: teller Kathleen Golli-her, for example, described the perpetrator as "Hispanic,” while Donovan Grierson, another bank customer, claimed he saw "a short Asian woman.”
. Jernigan points out that the getaway car during the October 11, 2000 robbery (for which she was originally charged but not tried) — a black Toyota 4-Runner SUV— matched the description of the getaway car used during the November 30, 2000 robbery (for which Rodríguez-Gallegos was charged). Moreover, although the victim from the October 11, 2000 robbery could not identify the robber in the photospread containing Jerni-gan's photograph, the victim, when shown the image from the September 20, 2000 robbery, declared that the person in that photograph was the person who robbed her. Although the majority discounts the significance of these details, they suggest that, in fact, Rodrí-guez-Gallegos — not Jernigan — was the perpetrator of the September 20, 2000 robbery.