In People v LeGrand (8 NY3d 449, 452 [2007]), the Court of Appeals ruled that
“where the case turns on the accuracy of eyewitness identifications and there is little or no corroborating evidence connecting the defendant to the crime, it is an abuse of discretion for a trial court to exclude expert testimony on the reliability of eyewitness identifications if that testimony is (1) relevant to the witness’s identification of defendant, (2) based on principles that are generally accepted within the relevant scientific community, (3) proffered by a qualified expert and (4) on a topic beyond the ken of the average juror.”
It is indisputable that this case turns on the accuracy of eyewitness identifications. Indeed, as the trial court noted, it is the only issue. It is also indisputable that there was no evidence other than the testimony of these eyewitnesses to connect defendant to the crime. What is disputed is whether the multiple eyewitness identifications in this case can serve as corroborat*182ing evidence. The plurality believes that the eyewitnesses can corroborate each other, thereby rendering an expert unnecessary under LeGrand. Because, under the facts of this case, misidentification remains a risk, the court’s refusal to permit any expert testimony constituted an abuse of discretion under LeGrand, and, more recently, People v Abney (13 NY3d 251 [2009]). Therefore, I must dissent.
The crime occurred on January 10, 2003 sometime before 5:30 a.m. at the Brooklyn Bridge subway station on the northbound J subway platform. The victim and her assailant were strangers. The assailant approached the victim, a 21-year-old female student. He had a brief conversation with her, then took out a box cutter and proceeded to slash her on the right side of her head and her left hand. The victim provided police with a description of her assailant immediately after the attack. Following a police investigation, the victim identified defendant in a photo array on January 24, 2003 and in a lineup on January 26, 2003.
There were two other eyewitnesses to this crime. Pablo Alar-con was on the J train platform on his way to work and made eye contact with the assailant who walked by with his face partly covered. Although a hood partially covered the assailant’s face, Alarcon took note because the assailant had a “suspicious” expression that made Alarcon feel afraid. A short time later, Alarcon heard a woman scream and turned to the right to see the assailant attacking a woman who was trying to defend herself. The assailant walked away from the attack, passing Alarcon, who saw him put away a type of knife into his pocket. The assailant escaped into the subway tunnel.
Despite his relatively close look at the perpetrator, Alarcon did not pick defendant out of a photo array. Nor did Alarcon inform the police that he recognized defendant in the January 26 lineup, even though he was “eighty percent” certain that he did. Alarcon initially lied to police, he said, because his immigration status made him afraid. Alarcon subsequently recognized a picture of defendant in the newspaper as the perpetrator. It was not until December 2003, 11 months later, that Alarcon admitted to the District Attorney’s office that he had actually recognized defendant in the lineup, but had lied to police. He was subsequently shown a photograph of the January 26, 2003 lineup. At that point he identified defendant, with 80% certainty, as the man he had recognized earlier, but had chosen not to identify. At the suppression hearing, Alarcon also identified defendant as the perpetrator.
*183Another eyewitness, a man named Edwin Rios, was descending the stairs on the subway platform. He walked past the victim and the assailant while they were conversing. He could see the assailant’s entire face and noted that he had a dark goatee and wore a cap, jeans and a hood. Rios also saw the assailant’s face again as the assailant ran away from the attack. On January 13, 2003, Rios confirmed to police the accuracy of the sketch a police artist had prepared with assistance from the victim. However, it was not until January 20, 2004, a few days prior to the commencement of trial, that Rios identified defendant at another lineup. At trial, Rios identified defendant as the attacker.
Prior to the trial, defendant sought to introduce expert testimony from Professor Steven Penrod1 on 12 subjects related to the accuracy of witness identifications. Specifically, defendant asked to elicit Penrod’s expert opinion about: (1) exposure time; (2) cross-racial identification accuracy; (3) similarity of lineup fillers increasing the accuracy of identification; (4) how lineup instructions can affect the eyewitness’s willingness to make an identification; (5) rate of memory loss; (6) influence of information acquired after the event; (7) wording of witness questioning; (8) unconscious transference to crime scene of person seen elsewhere; (9) preexisting witness attitudes and expectations; (10) simultaneous versus sequential lineups; (11) weapon focus; and (12) factors influencing witness confidence. In opposition, the People argued that the proffered testimony should be excluded as irrelevant, within the average juror’s ken and not generally accepted within the scientific community. Alternatively, the People sought a Frye hearing to determine whether Professor Penrod’s testimony would be based on professionally reliable sources.
In a written decision dated September 23, 2003, the motion court (Budd Goodman, J.) ordered a Frye hearing, to be conducted by the trial court, on the issue of whether defendant’s proffered expert testimony was based on principles generally accepted in the scientific community and would be relevant to the case. Justice Goodman wrote that the trial court was “best positioned to determine whether the expert’s testimony is ‘generally accepted’ and whether the proffered testimony is relevant and not unduly prejudicial.” In a subsequent hearing conducted on December 9, 2003, Justice Goodman reiterated *184that it was for the trial judge to decide which, if any, of the 12 proposed subjects for expert testimony were appropriate.
However, on December 19, 2003, the trial court (McLaughlin, J.) denied both defendant’s request for expert witness testimony and a Frye hearing. Citing People v Lee (96 NY2d 157 [2001]), Justice McLaughlin held that “this is not an appropriate case for an expert witness on any aspect of identification testimony. Consequently, the court will not conduct a Frye hearing as previously ordered” (2 Misc 3d 652, 653 [2003]). He reasoned that expert testimony concerning accuracy of identification was unnecessary because the victim had ample opportunity to observe her attacker, provided a description immediately after the attack, helped a police artist render a composite sketch one day later, selected defendant’s photograph from an array two weeks thereafter, and, on the next day, identified defendant in a lineup (id.).2
More specifically, Justice McLaughlin ruled that the proposed topic of weapon focus was not relevant because the victim never saw a weapon (id. at 654), that expert testimony as to how police investigative procedures might affect a lineup would be inappropriate as the victim would naturally assume that the person she had identified in the photo array would appear in the following day’s lineup (id.) and that the composite sketch prepared the day after the assault memorialized the victim’s recollection of her attacker, negating the potential significance of postevent information, the “forgetting curve,” the wording of questions and confidence malleability (id.). Justice McLaughlin also ruled that there were no generally accepted scientific principles concerning postevent information and unconscious transference (id. at 655), that issues concerning sequential lineups are not appropriate for jury resolution, that defendant and the victim are not members of different races as contemplated in the research literature and that the issue of the victim’s confidence vis-á-vis her identification could be sufficiently addressed in voir dire and the jury charge (id.).
At trial, after resting his case and the denial of his motion to dismiss the indictment, defendant renewed his motion to admit expert testimony on identification. The court denied that request also.
*185Notably, at the time he made his rulings, Justice McLaughlin did not have the benefit of the Court of Appeals decision in People v LeGrand (8 NY3d 449 [2007], supra), because that case was decided several years after the trial in this case. However, LeGrand still applies (see People v Vasquez, 88 NY2d 561, 573 [1996]). As stated earlier, in LeGrand, the Court of Appeals held that it is an abuse of discretion for the trial court to preclude expert testimony where a case turns on the accuracy of eyewitness identifications (8 NY3d at 452).
The police initially derived the LeGrand defendant’s identity from a composite sketch based on information from five witnesses to a murder. It was not until two years after the crime that a police officer first concluded that the sketch resembled the defendant, who was under arrest for an unrelated burglary (LeGrand, 8 NY3d at 452-453). The five witnesses did not view a photo array or a lineup until seven years after the crime, at which time only one of them was able to identify the defendant positively from a photo array and at a lineup (id. at 453). Two of the remaining witnesses were shown a photo array. One witness picked out the defendant’s photo as a “close, if not exact” match, while the other described the defendant’s picture as “similar” to that of the assailant (id.). The fourth and fifth witnesses were unable to identify the defendant (id.).
The Court of Appeals held that the trial court abused its discretion by precluding expert testimony on eyewitness identification, particularly with respect to three factors, all of which are at issue here: (1) correlation between confidence and accuracy of identification, (2) the effect of postevent information on accuracy of identification, and (3) confidence malleability (8 NY3d at 458). The Court held that these factors were relevant to the facts and issues of that case (id. at 456-457), and that the defendant had established that they were generally accepted in the scientific community (id. at 458). The Court noted, however, that “not all categories of [eyewitness expert] testimony are applicable or relevant in every case,” and that “admissibility [under LeGrand] would rest within the trial court’s sound discretion” (id. at 459).
LeGrand followed on the heels of People v Young (7 NY3d 40 [2006]). In Young, as in this case, the assailant’s face was partially covered. He entered the victims’ home wielding an axe and sledgehammer and threatened to kill one of the residents who was in a wheelchair. The assailant remained in the victims’ presence for five to seven minutes until the victims turned over *186their cash. The assailant also stole several watches. The victims later found that other property, including binoculars and a pair of gloves, was missing from their cars. Calling the question “close” (id. at 42), the Court of Appeals held it was not an abuse of discretion to deny the admission of expert testimony concerning the accuracy of eyewitness identification. However, the Court noted that
“if this case turned entirely on an uncorroborated eyewitness identification, it might well have been an abuse of discretion to deny the jury the benefit of [the expert’s] opinions.
“The corroborating evidence, however, significantly diminishes the importance of the proffered expert testimony in this case” (id. at 45-46).
That corroborating evidence consisted of the police finding the stolen property in the possession of two of defendant’s female acquaintances, one of whom claimed to have received the items from the defendant.
Quite recently, in People v Abney (13 NY3d 251 [2009]), the Court of Appeals reinforced that LeGrand's holding is not restricted to its facts. In Abney, the victim was mugged and a gold necklace was taken at knifepoint in a subway station. No evidence other than the victim’s identification connected the defendant to the crime. Before jury selection, the defendant made a motion in limine for an expert to testify about the accuracy of eyewitness identification. The trial court denied this motion. The defendant renewed his motion, on a narrower basis, at the close of the People’s case. The trial court denied this too. This Court affirmed the subsequent judgment of conviction and sentence (57 AD3d 35 [2008]). The Court of Appeals reversed, holding that, by the time the People’s case closed, “it was clear that there was no evidence other than [the victim’s] identification to connect defendant to the crime, and she did not describe him as possessing any unusual or distinctive features or characteristics” (13 NY3d at 268).
Here, as in Abney and LeGrand, there was no corroborating evidence and, as Justice McLaughlin noted, the case turned purely on eyewitness testimony (see 2 Misc 3d at 653 [identification the “only issue”]). Moreover, misidentification was a risk. None of the eyewitnesses described the perpetrator as possessing any unusual physical features, perhaps because a hood partially covered the perpetrator’s face. None of the witnesses *187were acquainted with defendant prior to the attack. The attack was brief and the perpetrator escaped into the subway tunnel. Therefore, LeGrand’s plain language requires, with regard to each category of proffered expert testimony, a trial court to determine whether that evidence would be (1) relevant to the witness’s identification of defendant, (2) based on principles that are generally accepted within the relevant scientific community, (3) proffered by a qualified expert and (4) on a topic beyond the ken of the average juror (8 NY3d at 452). Without the benefit of the LeGrand decision for guidance, that analysis did not occur in this case. Rather, the court believed the risk of misidentification to be low because the victim had the opportunity to observe defendant, had helped with the sketch and had picked him out of a photo array and then a lineup. The court then decided that an expert was unnecessary or that the proposed topics were within the common understanding of the jury.
With respect to several of the topics defendant proffered, the court was correct to conclude that an expert was unnecessary on the particular facts of this case. For instance, the court’s analysis was probably correct with respect to how the similarity of fillers influences accuracy of identifications because the victim had previously identified defendant in a photo array and should have realized that the person she selected would be in the lineup. The same analysis would also apply to the topic of lineup instructions. Nor does defendant contest the People’s arguments concerning: (1) exposure time, (2) cross-racial inaccuracy, (3) wording of investigator questions and (4) unconscious transference. Accordingly, defendant has abandoned on appeal his efforts to introduce expert testimony on these topics.
However, the court abused its discretion by not allowing expert testimony on the remaining topics. These subjects are relevant given the particular circumstances of this case. For example, Dr. Penrod was to testify about the “forgetting curve,” i.e., that memory loss is greater directly after an event. While it is common knowledge that memory fades over time, it is not well known, indeed it is counterintuitive, that the greatest memory loss occurs directly after an event. This knowledge is therefore outside the ken of the average juror (see Abney, 13 NY3d at 259, 268). Given that the victim claimed to remember her assailant well enough to assist a police sketch artist, and the trial court’s heavy reliance on the sketch to reason that an expert was unnecessary, the issue was certainly relevant. At the very least, the court should have held a Frye hearing to *188determine whether the theory behind the “forgetting curve” is generally accepted in the scientific community.
Principles about witness confidence are already generally accepted within the scientific community (id. at 268). “They are also counterintuitive, which places them beyond the ken of the average juror” (id.). The issue of witness confidence was relevant because two of the witnesses were certain defendant was the perpetrator, while the third was nearly certain. The jury therefore deserved to hear testimony that confidence level is not necessarily related to the accuracy of the identification and that certain factors, such as trial preparation, can affect witness confidence. The existence of the sketch has nothing to do with the tenor of the witness’s testimony and therefore should not have been a basis for precluding expert testimony about witness confidence. Finally, the court’s alternative method of addressing the issue of witness confidence during voir dire or charging, “as it always does,” that confidence does not constitute accuracy (2 Misc 3d at 655) is insufficient to dispel the risk of false identification.
Nor was the court correct to conclude that the sketch obviated the need for expert testimony. Nothing in this record supports a finding that the inaccuracies involved in eyewitness identification are ameliorated when a victim can assist a police sketch artist. Defendant never had the opportunity, despite making the request, to have his expert address this subject.
As it is undisputed that this case turned on the accuracy of eyewitness identification testimony and there was no evidence to corroborate that testimony, the court abused its discretion under LeGrand. Accordingly, defendant should be entitled to a new trial after a hearing on his offer of expert testimony during which the court should undertake the analysis LeGrand sets forth.
Nor do I consider the court’s failure to conduct this hearing to be harmless. For a nonconstitutional trial error to be harmless:3 (1) proof of guilt must be overwhelming; and (2) there must be “no significant probability that the jury would have acquitted had the proscribed evidence not been introduced” (People v Kello, 96 NY2d 740, 744 [2001]). As the trial court recognized, the only issue in this case was the identification of *189defendant. Yet, this case involved a real risk of misidentifications. None of the witnesses had an extended opportunity to view the assailant whose face a hood partially covered. None of them had ever seen him before. The attack was brief and the perpetrator escaped into the tunnel. No one followed him. Nothing of the victim’s was found with defendant. Certainly then, the proof of defendant’s guilt was not “overwhelming.” Had the court permitted defendant to challenge the reliability of the identifications by means of expert testimony (proffered to include testimony relating to the “forgetting curve” and witness confidence) it is entirely possible the jury would not have believed the accuracy of the identifications. Therefore, I cannot agree with the plurality that the three eyewitnesses (victim included) were “unlikely to be mistaken,” and therefore conclude that the trial court’s error in excluding expert testimony was not harmless.
Despite the clear mandate from the Court of Appeals to use the LeGrand factors where a case turns on eyewitness testimony and there is little or no corroborating evidence, the plurality continues to interpret LeGrand narrowly, in an attempt to restrict its reach by corralling its facts. The plurality mentions not only the lack of forensic or other physical evidence linking the defendant in LeGrand to the crime, but also the seven-year interval in that case between the commission of the crime and the witness identifications. The plurality then follows a series of decisions from this Court that have distinguished LeGrand on the ground that the risk of misidentification was slight compared to the risk in LeGrand (e.g. People v Chisolm, 57 AD3d 223 [2008], lv denied 12 NY3d 782 [2009]; People v Smith, 57 AD3d 356 [2008], lv denied 12 NY3d 821 [2009]; People v Austin, 46 AD3d 195 [2007], lv denied 9 NY3d 1031 [2008]).4
However, in light of the Court of Appeals decision in Abney reversing this Court, it is now abundantly clear that we are not to restrict LeGrand to its facts, even if the risk of misidentification is only slight. The legal analysis is quite simple: if the case turns on eyewitness identifications and there is little or no corroborating evidence, the trial court abuses its discretion if it denies appropriate expert testimony on eyewitness identifications.
*190The plurality interprets the recent Court of Appeals decision in People v Allen, decided with Abney, to mean that “corroborating evidence. . . need not be forensic or physical, but can be established by additional eyewitness testimony.” However, merely because there may be more eyewitnesses than just the victim does not mean an expert is dispensable. LeGrand recognizes that there is now persuasive scientific evidence that, under certain circumstances, eyewitness identification testimony, even while apparently convincing and certain, is fraught with error (see LeGrand, 8 NY3d at 454-455; see also State v Delgado, 188 NJ 48, 60, 902 A2d 888, 895 [2006] [“Misidentifieation is widely recognized as the single greatest cause of wrongful convictions in this country”]). Without additional circumstances, such as prior familiarity with the defendant, perhaps from the neighborhood, a prior encounter with or even a more ample opportunity to observe the perpetrator, the problem of misidentifieation can exist whether there is one eyewitness or several. The Court of Appeals has recognized this. After all, LeGrand involved at least three eyewitnesses to a murder and the LeGrand rule speaks in terms of accuracy of eyewitness “identifications” in the plural, not singular. Therefore, LeGrand dictates that, without more, where a case turns on the accuracy of more than one eyewitness, an expert is still necessary. The Court of Appeals’ juxtaposition of the circumstances in Allen with Abney reinforces this. In Allen, the Court of Appeals held the case did not turn on the accuracy of eyewitness identification because the defendant was not a stranger to either the victim or the eyewitness. Allen involved a robbery at a barbershop. One of the customers recognized one of the robbers, whose face was partiálly covered, as someone whom he encountered regularly in the neighborhood. The customer had heard the defendant’s speaking voice several times during the previous six months and recognized him from both his voice and body type. The defendant also was not a stranger to one of the barbers who worked in the shop. The Court of Appeals noted that two witnesses recognized the defendant from prior encounters when it ruled that Allen was not a case that turned on the accuracy of eyewitness identification. Here, in sharp contrast, the perpetrator was a stranger to all involved. Nor does it make a difference that the witnesses may have noticed the perpetrator before the crime took place. That exposure was of short duration and the perpetrator’s face was partially covered. The crime took place quickly and the perpetrator escaped into the subway tunnel.
*191Nor can I agree with Justice McGuire’s analysis, despite its compelling articulation, that focuses on the single word “independently” from Allen to conclude that “multiple corporeal identifications that are ‘independent ]’ can cross-corroborate each other so that the case is not one in which there is but Tittle or no corroborating evidence.’ ” Allen involved more than a mere independent identification. As Justice McGuire acknowledges, the Court of Appeals also took into consideration that the defendant in Allen was not a stranger to either eyewitness and had stood by and searched one of them throughout the course of the robbery. Nothing close occurred in this case. The perpetrator did not stand by or search either Rios or Alarcon throughout the course of the crime. Rather, the perpetrator’s face was partially covered, the crime occurred quickly and the perpetrator escaped into the subway tunnel. More importantly, unlike in Allen, none of the eyewitnesses were acquainted with defendant prior to the crime. Finally, neither Rios or Alarcon identified defendant as the assailant until nearly a year after the crime occurred.
Nor do I consider the victim’s interaction with the perpetrator to constitute “an extended opportunity to observe her attacker.” Not only was the assailant’s face partially covered, but the conversation the two had was of very short duration. Indeed, the conversation here appears not to be much longer than the conversation the victim and the perpetrator had in Abney, yet the Court of Appeals still held in Abney that it was an abuse of discretion not to allow an expert to testify. In Abney, like this case, the crime occurred on the subway. The assailant stopped the victim on the stairs of the subway station and asked her whether she had any change. The victim told the assailant she had no change. The two parted ways. Then, the assailant suddenly came around in front of the victim and committed the crime.
Accordingly, while I agree with Justice McGuire that it is possible to conceive of circumstances where multiple corporeal identifications can cross-corroborate each other, such as the eyewitnesses having prior acquaintance with the defendant or the defendant possessing a tattoo, scar or other distinguishing feature, this is not that case. The risk of misidentification here is too high. LeGrand therefore required the trial court to allow expert testimony to enhance the quality of the deliberative process as well as the reliability of its outcome.
*192Saxe, J.E, concurs with Buckley, J.; McGuire, J., concurs in a separate opinion; Moskowitz and Acosta, JJ., dissent in a separate opinion by Moskowitz, J.
Judgment, Supreme Court, New York County, rendered April 28, 2004, affirmed.
. The People do not contest Dr. Penrod’s qualifications.
. In his ruling, Justice McLaughlin made factual findings only with regard to the victim’s identification, because, at this point pre-trial, neither Rios nor Alarcon had yet identified defendant.
. Because I believe that the error was not harmless under the standard applicable to nonconstitutional trial errors, I take no position with respect to whether the Constitution requires or precludes a harmless error analysis in this case.
. In Chisolm and Smith there was evidence beyond that of the eyewitnesses to corroborate that the defendants were the perpetrators. Austin, however, had no such evidence and is inconsistent with LeGrand.