OPINION

DAVIS, Circuit Judge:

Appellant Deshawn Greene appeals convictions of armed bank robbery, 18 U.S.C. §§ 2, 2113(a), (d), and brandishing a firearm during and in relation to a crime of violence, 18 U.S.C. §§ 2, 924(c)(1)(A)(ii), for which he received consecutive sentences totaling 30 years in prison. At trial, through a series of leading questions to which no objections were made, the government elicited so-called "resemblance testimony" from a bank teller who had made no out-of-court identification and concededly could not make an in-court identification of Greene as the robber. Furthermore, in the absence of a request by the defense, the district court failed to give a Holley-Telfaire instruction.1 Greene argues on appeal that we should find plain error and award him a new trial on the basis of these circumstances. For the following reasons, we affirm the judgment.

I.

A.

1.

On May 6, 2009, an armed individual robbed the Fifth Third Bank in Kannapolis, North Carolina. The individual entered the bank around 11:30 a.m., pointed a silver-colored revolver at two employees, and demanded money. The robber first walked up to the counter of teller Alice Bolder, who was so frightened that she got under her counter. He then turned toward teller Kevin Morrison, pointed the gun at Morrison's chest, and demanded money. Morrison emptied a cash drawer and put the money into a bag. The robber then returned his attention to Bolder, pointing the gun at her and telling her to get up. Bolder did so, and she placed cash, along with a dye pack, into a purple bag given to her by the robber. The robber then left the bank. The total amount taken was $1,798.

2.

Witnesses gave police varying accounts of the appearance of the robber, who was wearing a disguise, in the immediate aftermath of the event, and they later testified to varying descriptions at trial. Shortly after the robbery, Bolder described the robber as an African-American male wearing a female wig, a long skirt, pants underneath the skirt, sneakers, a felt-type jacket with an emblem on it, large sunglasses, and carrying a purple tote bag. On one page of a robbery description form, Bolder described the robber as being 6-feet-5-inches tall and weighing about 160 pounds. On a second page of the form, she wrote that the robber was 6-foot-2. Morrison described the robber as a male wearing a long black skirt, a wig, large sunglasses, and a black hoodie. In a robbery description form, Morrison wrote that the robber was between 6 feet and 6-foot-2 and appeared to weigh between 140 and 160 pounds.

At trial, bank employee Kathy Jarvis testified the robber was an African-American "dressed in all black," but provided no further description. J.A. 74. She said she could not tell if the robber was a man or a woman. Jarvis testified that the robbery took three minutes. Morrison testified that it took five to ten minutes, and Bolder testified that it lasted ten to fifteen minutes.

A witness standing outside the bank, Sonya Shell, testified that she saw a "strangely dressed" person with a red wig, but could not tell if the person was a man or a woman. J.A. 172-73. Shell testified that after the robber left the bank, she saw the dye pack explode — "a big cloud of pink smoke went up in the air" — and the robber got into a silver car and "they took off." J.A. 168.

3.

The investigation of the robbery that ultimately led to Appellant Greene first focused on the silver getaway vehicle. An anonymous tip alerted police that the car involved in the robbery could be found at a house in nearby Enochville, North Carolina. An officer went to the house and found a silver Honda belonging to Angela Lear. During a consent search of the vehicle, the officer noted red stains inside the car consistent with the discharge of a red dye pack, and later testing confirmed that the stains were consistent with substances contained in such packs. Officers located Angela Lear's husband, Jay Dustin Lear, who told police at first that he had loaned the silver Honda to a crack dealer named "Slim" on the day of the robbery. In a second interview soon thereafter, Lear changed his story. He admitted that he and Greene (known to Lear by his street name, "Train") planned the bank robbery and that Greene was the one who entered the bank.2

Specifically, Lear testified as follows, pursuant to a plea agreement after pleading guilty to his role in the robbery. On the morning of the robbery, he picked up Greene at the home he shared with his girlfriend. Greene had a chrome handgun. He dropped Greene off at the Fifth Third Bank, drove down the block, and then drove back to the bank to pick up Greene after the robbery. They then drove to a nearby apartment where they stashed the stolen money and the costume Greene wore in the robbery. Lear gave some of the money to two friends, who took it to car washes to "recycle" it by exchanging it for coins. J.A. 119. Later, police found red-stained United States currency in the car wash change machines.

Police found no physical evidence linking Greene to the crime. They searched for but did not find any identifiable prints at the bank. They did not find any of Greene's fingerprints in the silver Honda. Police did not process for fingerprints the currency recovered from the change machines. Police searched for but could not locate the articles used in the robbery — the purple bag, the dress, the wig, and the gun (although they recovered a pair of sunglasses from Greene's residence). Police never asked any witnesses to the robbery, including the bank employees, to identify any potential suspect in a lineup or photo array. Thus, the only direct evidence of Greene's participation in the robbery was Lear's testimony to that effect. Lear, a longtime drug addict and a convicted felon, was subjected to vigorous cross examination.

4.

On direct examination at trial, some seventeen months after the robbery, bank teller Bolder described the robber as a black male wearing a skirt, a wig, and sunglasses, but she was not asked to attempt an identification of Greene and she did not identify him as the robber. On cross examination, she said the robber was about 6 feet tall but acknowledged that on the robbery description form, she had described the robber as 6-foot-5. She also stated that she was never asked to identify the robber through a lineup or photo array.

Then, on redirect examination, the following exchange occurred between the prosecutor and Bolder:

J.A. 93-94. The defense did not object to this line of questioning.

B.

The jury found Greene guilty of bank robbery, 18 U.S.C. § 2113(a); armed bank robbery, 18 U.S.C. § 2113(a), (d); and brandishing a firearm during and in relation to a crime of violence, 18 U.S.C. § 924(c)(1)(A)(ii). The district court merged the bank robbery conviction into the armed robbery count and imposed a sentence of twenty-three years for armed bank robbery and seven years for brandishing a firearm, to be served consecutively, for a total of 30 years.

Greene noted a timely appeal to this Court. We have jurisdiction pursuant to 28 U.S.C. § 1291.

II.

Greene appeals his convictions on two grounds. First, he argues the district court erred in admitting Bolder's testimony, which he claims was the product of unnecessarily suggestive questioning that resulted in the admission of unreliable identification evidence violative of due process. Second, he argues the district court erred in not providing the jury with a Holley-Telfaire instruction regarding eyewitness identification.

As to both issues, our standard of review is plain error because Greene did not object at trial to the identification testimony or the lack of a Holley-Telfaire instruction. See Fed.R.Crim.P. 52(b) ("A plain error that affects substantial rights may be considered even though it was not brought to the court's attention."). Under the plain error standard, the appellant must show (1) there was error; (2) the error was plain under current law; and (3) the error affected appellant's substantial rights. United States v. Rolle, 204 F.3d 133, 138 (4th Cir.2000). Finally, for this Court to notice the error, the error "must seriously affect the fairness, integrity or public reputation of judicial proceedings." Id. (citation and internal quotation marks omitted). This Court must notice an error that "causes the conviction or sentencing of an actually innocent defendant." Id. at 139 (citation and internal quotation marks omitted).

III.

A.

We first consider whether the government erred in eliciting, and the district court erred in admitting, the testimony of bank teller Bolder.

We begin by examining the background of so-called "resemblance evidence" inasmuch as the government seems to contend that such evidence is not subject to the same level of constitutional scrutiny as more classic eyewitness identification evidence. This Court has viewed resemblance testimony with skepticism. In Patler v. Slayton, 503 F.2d 472 (4th Cir.1974), we stated that "if there is a line between `resemblance' and `identification' testimony it is admittedly thin. Although thin, we think it is a line worth drawing." Id. at 476 (citation omitted). But Patler differs from this case in important ways.

In Patler, a witness who, from her car, had seen a man appear in front of her and then run behind her, testified that the man had "dark hair," "a dark complexion," and "was of medium build and height." Id. at 474. At trial the witness was asked to describe the person she saw. Id. The following colloquy took place:

Id. The witness previously had twice been asked by the police to attend the defendant's preliminary hearings to get a better view of him to help her identify him. Id. We strongly condemned such a procedure but held that the witness's descriptions of the perpetrator were "less dangerous" than direct identification testimony, and thus affirmed the conviction. Id. at 477.

We note that, as egregious as the police tactics in Patler were, on the stand the witness was asked to describe the person she saw the day of the crime; her attention was not directed to the defendant seated in the courtroom. Id. Furthermore, without prompting by the prosecutor, she volunteered her opinion that the man she saw "looked something like" the defendant on trial. Id. Other cases in which courts allowed resemblance testimony are similar: Witnesses were asked to describe what they had seen, or if anyone in the courtroom resembled the perpetrator. See, e.g., United States v. Garcia-Ortiz, 528 F.3d 74, 79 (1st Cir.2008) ("At trial, the Government asked Gomez if anybody in the courtroom resembled the person that Rivera identified."); United States v. Bush, 749 F.2d 1227, 1231 (7th Cir.1984) (the witness was asked, "Do you see anybody here in the courtroom today that resembles one of the two men who robbed you on that day ...?") (ellipsis in original); United States v. Brooks, 449 F.2d 1077, 1081 n. 3 (D.C.Cir.1971) (witnesses were asked, "Do you see anybody/anyone in the courtroom today who resembles the person you saw?").

Here, the witness was asked not to describe the perpetrator, as in Patler, or if anyone in the courtroom resembled the perpetrator, as in the above cited cases; rather, the prosecutor pointedly drew the witness's attention to the defendant and instructed her to tell the jury what similarities existed between the defendant and the robber. See J.A. 93-94. The district judge, who was in the courtroom and best positioned to consider the testimony in its proper context, took it as identification testimony. See J.A. 97 ("I'm still thinking about this in-court identification.").

The Supreme Court has established a two-step process to determine whether identification testimony is admissible. See Manson v. Brathwaite, 432 U.S. 98, 110, 97 S.Ct. 2243, 53 L.Ed.2d 140 (1977); Satcher v. Pruett, 126 F.3d 561, 566 (4th Cir.1997). "First, the court must consider whether the identification procedure is unnecessarily suggestive." Satcher, 126 F.3d at 566. "Second, if the procedure was unnecessarily suggestive, a court must look at several factors to determine if the identification testimony is nevertheless reliable under the totality of the circumstances." Id.3 Those factors were set out by the Supreme Court in Neil v. Biggers, 409 U.S. 188, 199-200, 93 S.Ct. 375, 34 L.Ed.2d 401 (1972). Before turning to those factors, we first consider whether Bolder's testimony was unnecessarily suggestive.

1.

The Due Process Clause of the Fourteenth Amendment protects individuals from unreliable identifications that result from impermissibly suggestive procedures. See Manson, 432 U.S. at 113, 97 S.Ct. 2243. We have stated, "A procedure is unnecessarily suggestive if a positive identification is likely to result from factors other than the witness's own recollection of the crime." Satcher, 126 F.3d at 566. We have also noted that the phrasing of a question may suggest a desired response. Smith v. Paderick, 519 F.2d 70, 75 n. 6 (4th Cir.1975). For instance, the question, "What color tie was he wearing?" is not likely to elicit an answer of "None." Id. Without question, we are well aware of the danger of erroneous eyewitness identifications:

Id. at 75. We added, "Tainted identification evidence cannot be allowed to go to a jury because they are likely to accept it uncritically." Id.

The Second Circuit has found that when a defendant was the only African-American in the courtroom, and was seated at the defense table, the in-court identifications by three witnesses were "so clearly suggestive as to be impermissible." United States v. Archibald, 734 F.2d 938, 942-43 (2d Cir.1984). "Any witness, especially one who has watched trials on television, can determine which of the individuals in the courtroom is the defendant, which is the defense lawyer, and which is the prosecutor." Id. at 941. The Archibald court ultimately found, however, that admitting the in-court identifications was harmless error because the witnesses had also identified the defendant in photo arrays prior to trial. Id. at 943.

The Third Circuit held that when a witness sitting outside the courtroom saw the defendant walk past her in shackles and with a U.S. Marshal at each side, the witness's later in-court identification should not have been admitted. United States v. Emanuele, 51 F.3d 1123, 1130 (3d Cir.1995). The identification was "impermissibly suggestive" and obtained "in violation of defendant's right to due process." Id. at 1130-31. That witness, a bank teller, had been unable to identify the defendant in a photo array prior to trial. Id. at 1127. The court reversed the conviction because it determined that the in-court identification was "crucial evidence on the robbery's identity." Id. at 1132.

The Fifth Circuit, in another bank robbery case in which a teller's in-court identification was at issue, held that "it is obviously suggestive to ask a witness to identify a perpetrator in the courtroom when it is clear who is the defendant." United States v. Rogers, 126 F.3d 655, 658 (5th Cir.1997). In Rogers, when the teller first took the stand, she described what the robber was wearing. Id. at 657. But following cross-examination, the prosecutor thought that he saw something odd about the witness. Id. He asked an FBI agent to approach the witness, and she told the agent that she recognized the defendant as the robber. The witness was recalled to the stand and provided an in-court identification. Id. In holding the identification to be impermissibly suggestive, the Fifth Circuit noted that ten months had passed between the crime and the identification and stated that the circumstances surrounding the identification rose to the level of a due process violation. Id. at 659. "Even the best intentioned among us cannot be sure that our recollection is not influenced by the fact that we are looking at a person we know the Government has charged with a crime." Id.

The New Jersey Supreme Court last year undertook an exhaustive evaluation of eyewitness reliability in New Jersey v. Henderson, 208 N.J. 208, 27 A.3d 872 (2011). The court was particularly skeptical of identifications made as part of show-ups, which are similar to in-court identifications. Id. at 903. The court found show-ups to be "inherently suggestive," and stated:

Id. Reviewing the social science research, the court noted that show-ups performed within minutes of an encounter were just as accurate as lineups. Id. But reliability quickly declined. Show-ups occurring only two hours after the encounter frequently led to misidentifications. Id.

Manifestly, whether the testimony in the instant case is properly classified as resemblance testimony or identification testimony is not relevant to the suggestiveness inquiry. Our inquiry here concerns the questions asked by the prosecutor and the circumstances in which the witness offered testimony (on leading questions, no less) on the basis of which the jury could rest a finding (as corroborative of the direct testimony of an alleged accomplice testifying pursuant to a plea agreement) of the defendant's participation in the charged bank robbery.

The questioning here, and thus the circumstances leading to the identification evidence, was unnecessarily suggestive. It was clear who in the courtroom was the defendant. Bolder was asked to look at the defendant: "Can you look at him for a moment?" J.A. 93-94. The prosecutor then asked Bolder if she saw any similarities between the defendant and the bank robber. She said, "Yes," and the prosecutor instructed her, "Tell the members of the jury and the Court what those similarities are." J.A. 93-94. This is exactly the sort of suggestive questioning we warned about in Paderick, where the phrasing of a question suggests the desired response. As in Paderick, the witness here, who was herself a victim of the robbery, likely felt pressured to help solve a crime and understandably wanted to be of assistance. Pressed for similarities, she found some: "The nose, I remember the teeth, the slimness of the face, and vaguely the mouth." J.A. 93-94.

The suggestive nature of this line of questioning is as clear as it is impermissible. Sitting across the courtroom from the defendant, with the judge and jury looking on, and a prosecutor drawing her attention to the defendant and asking for similarities, the witness understandably may have felt pressure to find something in the defendant that reminded her of the bank robber. These circumstances present a suggestive situation in which it is not clear whether the witness's own recollections, or outside pressures, are driving the testimony. See Archibald, 734 F.2d at 941 (when the defendant is seated at the defense table throughout the trial, it is "obviously suggestive" to ask witnesses to make an in-court identification). We therefore hold the procedure used to obtain the testimony was unnecessarily suggestive.

We now turn to whether Bolder's testimony was reliable.

2.

Even if an impermissibly suggestive procedure is used to obtain an in-court identification, admission of the identification evidence is not error if the evidence was "nevertheless reliable under the totality of the circumstances." Satcher, 126 F.3d at 566; see also United States v. Wilkerson, 84 F.3d 692, 695 (4th Cir.1996). "The factors the court may consider in measuring reliability include: (1) the witness' opportunity to view the perpetrator at the time of the crime; (2) the witness' degree of attention at the time of the offense; (3) the accuracy of the witness' prior description of the perpetrator; (4) the witness' level of certainty when identifying the defendant as the perpetrator at the time of the confrontation; and (5) the length of time between the crime and the confrontation." Wilkerson, 84 F.3d at 695 (citing Biggers, 409 U.S. at 199-200, 93 S.Ct. 375).

Applying those factors to the instant case, we think the unreliability of the in-court identification was clear. First, we note that the witness's opportunity to view the perpetrator was limited. The parties agree that given the short period of time the robber was in the bank and that he was wearing a long wig and sunglasses, Bolder had little time in which to observe him. See Gov't Br. 21 ("With respect to the first factor ... the relatively short time at issue and the fact that the robber's face and body were both obscured by disguise indicate that Ms. Bolder's opportunity to view the robber was relatively limited.") (citations and internal quotation marks omitted). Similarly, the Fifth Circuit observed in a bank robbery case in which the robber was wearing wraparound sunglasses and a baseball cap: "In light of the short time for observation, and the extent to which the robber's face was obscured at the time of the crime, the witness's opportunity to view him must be regarded as relatively limited." Rogers, 126 F.3d at 658.

Second, Bolder's degree of attention to the robber at the time of the offense was greatly diminished due to her reasonable fear and the distraction of having a weapon pointed at her. Indeed, Bolder testified, "First instinct I did, I got under my counter because I was frightened." J.A. 83. She also said, "I didn't even hardly know my name that day; I was nervous.... I was fragile." J.A. 94. The Fifth Circuit in Rogers found that a witness's reasonable fear "does not change the fact that it weighs against the reliability of her identification by throwing some doubt on her ability to concentrate on and remember his face." 126 F.3d at 659. In the same vein, the New Jersey Supreme Court, in Henderson, noted, "Even under the best viewing conditions, high levels of stress can diminish an eyewitness' ability to recall and make an accurate identification." 27 A.3d at 904.

Reviewing the scientific research, the Henderson court also found that "weapon focus," when a weapon is visible during a crime, can affect a witness's ability to describe a perpetrator. Id. at 904-05. Weapon focus can "impair a witness' ability to make a reliable identification and describe what the culprit looks like if the crime is of short duration." Id. at 905. Here, Bolder had a gun pointed at her twice. She testified, "When he first approached my window, it was [pointed] towards my face; when I got up from behind the counter, towards my chest, towards my body." J.A. 85. This factor, then, must weigh against the reliability of her testimony.

Third, the accuracy of Bolder's prior descriptions of the robber weighs in favor of her reliability. While it is true that at various times Bolder gave different heights for the robber — on the robbery description form she stated the robber's height to be both 6-foot-2 and 6-foot-5, and at trial she testified the robber was 6 feet tall — it is also true that she described the robber's disguise in a way that closely aligned with the description given by her co-teller, Kevin Morrison. Both told police the robber wore a woman's wig, a skirt, and sunglasses. On these important details, as reflected in images of the robber captured by bank video cameras, Bolder was correct.

Fourth, the witness's level of certainty in identifying the perpetrator is a wash because while Bolder did not state with certainty that Greene was the bank robber, she did not equivocate in noting the similarities between Greene and the robber.4 The government conceded in its brief that Bolder's testimony "does not involve certainty that Defendant Greene was the robber." Gov't Br. 22.

Fifth, the length of time between the crime and the confrontation weighs against the reliability of the testimony. The government also concedes this point in its brief. See Gov't Br. 22-23 ("[T]he seventeen months between the robbery and the trial is a significant period of time."). During the seventeen months that passed between the bank robbery and Greene's trial, Bolder was not once asked to view a lineup or photo array or assist a police artist in drawing a sketch of the robber. The Supreme Court in Biggers stated that a lapse of even seven months between the crime and the identification "would be a seriously negative factor in most cases." 409 U.S. at 200, 93 S.Ct. 375. The Fifth Circuit in Rogers noted that a ten-month lapse "raises concerns about the accuracy of the memory." 126 F.3d at 659. Here, the seventeen months between crime and confrontation is an unquestionably lengthy period of time that must weigh against reliability.

Even while conceding that most of the Biggers factors militate against reliability, the government persists in arguing that Bolder's testimony should nevertheless be found reliable because "the non-`identification' evidence presented at trial was overwhelming in pointing to Defendant Greene as the robber of the Fifth Third Bank." Gov't Br. 24. But evidence extrinsic to an identification cannot be considered in evaluating the reliability of the identification. The Supreme Court made it clear in Manson that extrinsic evidence as to the reliability of an identification "plays no part in our analysis." 432 U.S. at 116, 97 S.Ct. 2243. Underlining the point, Justice Stevens in his concurrence in Manson approvingly noted that the majority "carefully avoids this pitfall and correctly relies only on appropriate indicia of the reliability of the identification itself." Id. at 118, 97 S.Ct. 2243 (Stevens, J., concurring). Justice Stevens stated that in evaluating the admissibility of identification testimony, courts must "put other evidence of guilt entirely to one side." Id. Extrinsic evidence may play a role in plain-error analysis (or, analogously, harmless error analysis), but it cannot be considered in assessing the reliability of Bolder's identification testimony.5

3.

In sum, the procedure used to obtain Bolder's testimony was suggestive, and unnecessarily so, because the prosecutor blatantly directed her to look at the defendant (after she testified she had intentionally declined to look at Greene during her entire time on the witness stand) and to describe similarities with the bank robber. The identification was also unreliable under the five Biggers factors. Bolder had a limited opportunity to view the robber, given the robber's disguise, his brief amount of time in the bank, and the presence of the firearm. Moreover, nearly a year-and-a-half elapsed between the robbery and the in-court identification. We therefore hold it was error to admit Bolder's testimony as to the similarities between Greene and the bank robber.

B.

For Greene to prevail under the plain error standard of review, the error must be plain. "An error is plain, at least, when the error is clear both at the time it occurred and at the time of appeal." United States v. Cedelle, 89 F.3d 181, 185 (4th Cir.1996). Like other courts, we have emphasized that in-court identifications are "the most dangerous evidence known to the law" because of "the very appreciable danger of convicting the innocent." Paderick, 519 F.2d at 75. The Supreme Court in Biggers, decided in 1972, stated that "[i]t is the likelihood of misidentification which violates a defendant's right to due process." 409 U.S. at 198, 93 S.Ct. 375. "Suggestive confrontations are disapproved because they increase the likelihood of misidentification, and unnecessarily suggestive ones are condemned for the further reason that the increased chance of misidentification is gratuitous." Id. In 1997, we reemphasized the two-step process for determining if identification testimony is admissible — asking first if the identification procedure was unnecessarily suggestive, and then if the identification was nevertheless reliable — and reiterated the five Biggers factors. Satcher, 126 F.3d at 566.

The government, hewing to its theme, argues that the law is not clear here because Bolder's testimony was resemblance testimony, not an identification, suggesting that such testimony does not merit the full panoply of due process protections. We disagree for the reasons we have already discussed. We agree with the District of Columbia Circuit, which noted in Brooks, 449 F.2d at 1083-84, that resemblance testimony merited due process protections. The Brooks court held that admitting resemblance testimony was harmless error, but error nonetheless, when there was "overwhelming" independent evidence of guilt. Id. The court stated, "While `resemblance' testimony projects some uncertainty on the part of the witness, it is part of the evidence which the jury may consider to constitute a basis for a guilty verdict, and a defendant's rights would be violated if such testimony had been obtained by the Government e.g., by an arrantly suggestive confrontation." Id. at 1083. And, as noted above, in Brooks, unlike in the instant case, the prosecutor asked if "anyone in the courtroom" resembled the perpetrator, id. at 1081, a significantly less suggestive procedure than the one employed here.

We are aware the Seventh Circuit held it was not plain error to allow resemblance testimony when a prosecutor asked a witness if she saw "anybody here in the courtroom today that resembles" one of the robbers. United States v. Bush, 749 F.2d 1227, 1231-32 (7th Cir.1984). The Bush court stated that "[w]hile this testimony may raise due process concerns, the impact of the testimony is not likely to be as great" as an identification. Id. at 1232. We question the continuing force of this observation in light of our deeper understanding, gained in the nearly thirty years since the trial in Bush, of the dangers posed by unnecessarily suggestive identification procedures. In any event, in Bush the prosecutor asked the witness to scan the entire courtroom; he did not direct the witness's attention to the defendant and ask her to describe similarities with the perpetrator, a tactic that increases the suggestiveness of the procedure and the likelihood of an irreparable misidentification.

It is well-settled that a prosecutor cannot verbally or physically point to a defendant and ask a witness if the defendant is the person who committed the crime. See United States v. Warf, 529 F.2d 1170, 1171 (5th Cir.1976) (holding that it was clearly "inadmissible and inappropriate" to point to the defense table while asking a witness if the perpetrator of a bank robbery was "in the courtroom"). The Warf court found the prosecutor's actions in pointing to the defendant to be so "dubious" that it reversed the conviction. Id. at 1174. Similar dangers lurk in a prosecutor's gratuitous elicitation by leading questions of damaging identification testimony in the guise of "resemblance testimony."

The law is plain: A prosecutor cannot point to the defendant, or direct the witness's attention to the defendant, and then elicit identification or resemblance testimony. We hold it was plain error to admit Bolder's testimony.

C.

Once we have determined that the district court committed error and that the error was plain, we must inquire whether the error "affected substantial rights." United States v. David, 83 F.3d 638, 646-47 (4th Cir.1996). "[T]he phrase `affecting substantial rights' in most cases ... means that the error [was] prejudicial." Id. (citation and internal quotation marks omitted) (ellipsis and insertion in original). "[W]hen considering whether the error was prejudicial, we look to the totality of the circumstances including all of the evidence adduced." Rogers, 126 F.3d at 659. The District of Columbia Circuit has held that it is harmless error to admit weak resemblance testimony when there is "strong evidence independent of the testimony of resemblance witnesses." Brooks, 449 F.2d at 1083. And we have held that an in-court identification made after the witness had observed the defendant at the defense table was, if anything, harmless error when there was "very strong" physical evidence linking the defendant to the offense. Satcher, 126 F.3d at 567, 569 n. 3.

Here, although it is a close question, the strong independent evidence linking Greene to the bank robbery persuades us that the error in admitting Bolder's testimony did not affect Greene's substantial rights. We note that Greene's accomplice, Lear, testified that he and Greene planned the robbery and that Greene entered the bank while Lear drove the getaway vehicle.6 Lear also testified that Greene possessed a chrome-colored gun, similar to the one used in the robbery. Greene argues that Lear's testimony is not credible because of Lear's admitted drug use and prior convictions. The credibility of a witness, however, is a task the law assigns to the jury, not appellate judges. See United States v. Cecil, 836 F.2d 1431, 1442 (4th Cir.1988) ("Credibility ... is for the jury — the jury is the lie detector in the courtroom.") (quoting United States v. Barnard, 490 F.2d 907, 912 (9th Cir.1973)). Further, Lear was subjected to a piercing cross examination.

The jury was well aware of Lear's credibility issues. He testified to his heroin addiction and that, indeed, he used heroin immediately after the commission of the robbery (rather than before the robbery) because he was able to make a purchase using the robbery proceeds. He admitted that years of illegal drug use had affected his long-term memory, that he had numerous felony convictions, and that he hoped his testimony would lead to a reduction in the time he would serve for his role in the robbery. The jury had all of this information. The determination of Lear's credibility was for the jury to make, not us.

In addition to Lear's directly inculpating testimony, other independent, though circumstantial, evidence of Greene's criminal agency is in the trial record. Janice Hester, Greene's girlfriend, testified that on the morning of the robbery, Lear came to her house to pick up Greene and they left the house together around 9:30 a.m. The robbery was committed at approximately 11:30 a.m. Furthermore, the government introduced into evidence photos of the robbery taken from a bank security camera and a photograph of Greene's right hand. The bank photos show scarring on the robber's right hand that is highly similar to the scarring in the photo of Greene's right hand.7 See also supra n. 6.

We therefore hold that because Bolder's testimony was "a final brush stroke rather than the essential outline of the picture identifying the [robber]," Satcher, 126 F.3d at 569, the error in admitting her testimony did not affect Greene's substantial right to a fair trial. Accordingly, we decline to exercise our discretion to grant relief on the basis of plain error.

IV.

Finally, Greene argues it was plain error for the court not to give the jury a Holley-Telfaire instruction. We have held that a Holley-Telfaire instruction on eyewitness identification should be given in cases where there is "no evidence of identification except eyewitness testimony." United States v. Holley, 502 F.2d 273, 275 (4th Cir.1974). Such an instruction advises the jury on how to appraise a witness's identification testimony, emphasizing whether the witness had adequate opportunity to observe the offender, how far the witness was from the offender, how good the light was, the length of time between the offense and the identification, and other factors. Id. at 277. We have cautioned that the Holley-Telfaire rule is a flexible one and not a rigid requirement on trial courts. United States v. Brooks, 928 F.2d 1403, 1408 (4th Cir.1991). "The Holley-Telfaire instruction or its substantial equivalent is not required to be given, sua sponte, in a case where other independent evidence, whether direct or circumstantial, or both, is presented to the trier of fact which is corroborative of the guilt of the accused." United States v. Revels, 575 F.2d 74, 76 (4th Cir.1978).

Because, assuming that Bolder's testimony should be treated as eyewitness testimony, there was independent evidence of Greene's participation in the robbery, we hold the district court did not err in failing to give a Holley-Telfaire instruction when such an instruction was not requested by the defense.

V.

We hold that the government's effort to craft a freestanding "resemblance testimony" carve-out from settled eyewitness evidence jurisprudence is unavailing, and that the government's examination of Alice Bolder under the circumstances here resulted in the elicitation of unnecessarily suggestive evidence of identification wholly lacking reliability. Nevertheless, for the reasons set forth, we decline to find reversible error on this record. We further hold the district court did not err in failing to give a Holley-Telfaire instruction. Accordingly, the judgment is

AFFIRMED.