OPINION OF THE COURT
Smith, J.Defendant was convicted of manslaughter in the first degree for causing the death of a man by holding him in a headlock. The duration of the headlock was an important issue at trial. Defendant argues that several of the trial court’s rulings in admitting and excluding evidence related to that issue were mistaken. As to one of those rulings—the court’s refusal to permit defendant to refresh his witness’s recollection with a statement the witness had previously given—we agree with defendant, and order a new trial.
I
The victim, Andrew Reister, was a bouncer in a bar. On the night in question, defendant and a young woman were in the bar, dancing on a table. Reister asked defendant to get off the table, defendant refused, and Reister pushed him off. There followed a fight. In short order, defendant got behind Reister and put his arms around his neck; one of defendant’s hands was grasping the other. After an interval, Reister fell to the floor and defendant fell on top of him, not releasing his grip, though Reister seemed to onlookers to be unconscious. Several people screamed at defendant to let Reister go, and some tried without success to pull defendant away. Finally, defendant let go and ran *374out of the bar, leaving Reister unconscious on the floor. Reister was declared brain dead two days later.
Defendant was indicted for murder and relied on a defense of justification (self-defense). At his trial, the People asked seven of their witnesses to estimate the duration of the headlock. The estimates varied, but most put the total time, beginning when defendant’s arms first went around Reister’s neck and ending when he released him, at somewhere near three minutes. Two defense witnesses gave shorter estimates; by their telling, the headlock may have lasted less than a minute.
The jury acquitted defendant of murder, but convicted him of manslaughter in the first degree (causing death with the intent to cause serious physical injury [Penal Law § 125.20 (1)]) as a lesser included offense. The Appellate Division affirmed (People v Oddone, 89 AD3d 868 [2d Dept 2011]). A Judge of this Court granted leave to appeal (20 NY3d 1102 [2013]), and we now reverse and order a new trial.
II
Of the issues raised by defendant on this appeal, we find three—all related to what witnesses were or were not allowed to say about the duration of the headlock—that call for discussion. Defendant challenges the following evidentiary rulings:
(1) James Wilson, the doctor who performed an autopsy on Reister’s body, was permitted to testify that in his opinion Reister’s neck had been compressed for “something in the range of 2, 3, 4 minutes.”
(2) When Megan Flynn, a defense witness, testified that the duration of the part of the incident she observed “could have been a minute or so,” defense counsel was not allowed to refresh her recollection with a prior statement that put the same interval at “maybe 6 to 10 seconds.”
(3) Steven Penrod, an expert in eyewitness observation, was not permitted to testify that eyewitnesses routinely overestimate, by a large margin, the duration of relatively brief events.
We reject defendant’s attack on Wilson’s testimony. We agree with defendant that the restriction placed on his questioning of Flynn was error requiring a new trial. Whether the exclusion of Penrod’s testimony was an abuse of discretion is a close question that we do not need to decide, but we offer some observations about it for the guidance of the court at a retrial.
*375A. Wilson
Wilson, a deputy medical examiner, inferred a 2-4 minute duration for the headlock principally from two facts: his own observation at the autopsy of “petechiae”—red spots caused by bursting of blood vessels—on and around Reister’s eyes; and the observations of several witnesses that, by the time the incident ended, Reister’s face had turned purple. As to the petechiae, Wilson testified:
“Q. Could you tell us, Doctor in your experience how long it would take for this type of petechia to be present in Mr. Reister’s—around his eyes, in the skin surrounding his eyes?
“A. Well, in my experience and understanding of how this process occurs an injury of this sort would take matter of a few minutes, 2, 3 perhaps 4, with neck compression on type some kind of a struggle. So there may be slight variations in the pressure from time to time, but matter of a few minutes, something in the range of 2, 3, 4 minutes.”
Similarly, as to the discoloration of Reister’s face, Wilson testified:
“Q. In your opinion, Doctor, how long would it take for the blood in the veins that is not able—that is being squeezed and kept in the head, how long would it take in order for that purple cast or coloration to occur in Mr. Reister’s face?
“A. Well, in my opinion and experience the blood that is built up over a period of time, then loss of oxygen, to get very dark it would be a matter of a few minutes minimum, something in the order of 2, 3, 4 minutes.”
Defendant attacks this testimony as lacking a scientific basis. He does not dispute that petechiae and purple coloring can result from neck compression, but he says—and the People do not dispute—that no scientific studies have been published to show how lengthy a compression is required to produce those results. Thus, defendant argues, Wilson was advancing a scientific principle that had not gained general acceptance in its field, in violation of the rule of Frye v United States (293 F 1013 [DC Cir 1923]), which is followed by the courts of New York (People v Wesley, 83 NY2d 417 [1994]).
*376The flaw in defendant’s reasoning is that Wilson did not claim to rely on any established scientific principle. He made clear that his testimony was based on his personal “experience”—meaning what he had observed, heard and read about particular cases. Such evidence is not barred by Frye (see Johnson v State, 933 So 2d 568, 570 [Fla App 2006] [“An expert opinion based on personal training and experience is not subject to a Frye analysis”]; Commonwealth v Devlin, 365 Mass 149, 155, 310 NE2d 353, 357 [1974] [“Dr. Sosman’s medical opinion . . . was not the product of a ‘scientific theory’ but was, rather, the product of years of experience”]).
Defendant argues in substance that an expert who is a scientist can express no opinion based on his own experience, but must rely only on published studies or texts. We reject the argument. It is true that an opinion based on experience alone is ordinarily less reliable than one based on generally accepted science. An expert may well overvalue his own experience, or even exaggerate or fabricate it. But these flaws can be exposed by cross-examination, and by the opinions of opposing experts—as the alleged flaws in Wilson’s testimony were in this case. There will ordinarily be no unfairness as long as the jury is not misled into thinking that the expert’s opinion reflects a generally accepted principle (see Flanagan v State, 625 So 2d 827, 828 [Fla 1993] [an expert’s reliance on “some scientific principle or test . . . implies an infallibility not found in pure opinion testimony”]).
We acknowledge that it may not be possible to draw a neat line between scientific principles and experience-based testimony. Indeed, it has been observed that the many cases applying Frye to evidence based on scientific principles shed little light on exactly what a “scientific principle” is (see 22 Charles A. Wright & Kenneth W. Graham, Jr., Federal Practice and Procedure: Evidence § 5168.2 [2d ed Apr. 2013]). We do not imply that an expert is allowed to say anything he or she likes to a jury if the statement is prefaced by the words “in my experience.” To allow an expert to say, based only on his or her alleged experience, that smoking does not cause lung cancer or that baldness is related to the phases of the moon would be to tolerate the admission of junk science and to undermine the basic purpose of Frye.
But Wilson’s testimony in this case does not trigger a concern of that kind. The parties here appear to agree that petechiae and discoloration are caused by neck compression; how long the *377neck must be compressed is a question that scientific studies do not seem to have answered. To allow a pathologist who has examined many dead bodies, and heard and read many accounts of how victims met their deaths, to express an opinion on the subject accords with common sense, and does not open the door to every expert’s flight of fancy.
B. Flynn
Flynn, a waitress at the bar where the fatal event took place, saw part of the incident and later told an insurance company investigator that the part she saw lasted “for maybe 6 to 10 seconds.” The People interviewed her before trial but decided not to call her as a witness. The defense did call her, and asked essentially the same question the insurance investigator had asked: “[F]rom the time that you walked in to the time you saw the guy let go how long of a period of time was it?” On the witness stand, Flynn gave a different answer: “I didn’t have a watch. I wasn’t keeping track of time. But it could have been a minute or so. I don’t know.” Defense counsel tried to show Flynn her previous statement to refresh her recollection, but was not permitted to do so. The trial court ruled that Flynn had “given no indication she needs her memory refreshed.”
In this, the trial court erred. When a witness, describing an incident more than a year in the past, says that it “could have” lasted “a minute or so,” and adds “I don’t know,” the inference that her recollection could benefit from being refreshed is a compelling one. More fundamentally, it was simply unfair to let the jury hear the “a minute or so” testimony— testimony damaging to the defense, from a defense witness’s own lips—while allowing the defense to make no use at all of an earlier, much more favorable, answer to the same question. The trial court suggested to defense counsel that this was “an effort to impeach your own witness,” but counsel had not yet got to the point of impeachment; she only wanted to refresh the witness’s recollection. And in any event, technical limitations on the impeachment of witnesses must sometimes give way, in a criminal case, to a defendant’s right to a fair trial (Chambers v Mississippi, 410 US 284 [1973]).
Though Flynn was certainly not the central witness in the case, we conclude that the error in limiting counsel’s examination of her was important enough to justify reversal. Indeed, the People do not argue that the error was harmless. It may be that her original “6 to 10 seconds” statement, if repeated on the *378stand, would have been of little consequence, for Flynn saw only the last part of what occurred. But the testimony of a defense witness that that fragment of the event might have lasted as long as a minute gave significant support to the People. The prosecutor used—indeed, overstated—Flynn’s testimony in her closing argument: “Megan Flynn even told you, the defense’s own witness, told you it was one to two minutes.”
C. Penrod
Defendant sought to call Penrod, a psychology professor, “as an expert on the issue of eyewitness observations,” explaining in a detailed offer of proof, accompanied by an affidavit from Penrod, what he would testify to. Much of his proposed testimony was on matters within the ken of the average juror, and the trial court was plainly right to exclude it. But one point that Penrod proposed to make, relating to the accuracy of estimates of duration, cannot be put aside so readily. Penrod said in his affidavit: “It is generally accepted in the field of forensic psychology that eyewitnesses routinely overestimate the duration of relatively short events lasting a few minutes or less.”
Penrod offered specific support for this assertion, quoting another psychologist who had studied the topic:
“Theoretical and empirical investigation of duration estimations date back to the nineteenth century with Vierordt’s (1868) discovery that short intervals tend to be overestimated and longer ones underestimated. This finding is now commonly referred to as ‘Vierordt’s Law’. Consistent with this law, forensically related research has shown that witnesses tend to overestimate the duration of relatively short events lasting a few minutes or less .... Yarmey and Yarmey (1997) found that witnesses in field situations (involving) ... a 15-second interaction with a ‘culprit’ . . . overestimated this encounter by a 3 to 1 ratio.”
Citing People v LeGrand (8 NY3d 449 [2007]), defendant argues that the exclusion of expert testimony about “Vierordt’s Law” was error. LeGrand, as the trial court observed, is not directly in point. The issue in LeGrand was the reliability of an eyewitness’s identification of defendant as the person who committed the crime, and we held it an abuse of discretion to *379exclude expert testimony about the reliability of such identifications where the case “turned solely on the accuracy of the witnesses’ identification,” for which there was “no corroborating evidence” (8 NY3d at 457). Here, defendant’s identity was never in issue.
LeGrand, however, can be read to stand for the broader principle that there are cases in which it is unfair to deprive the jury of expert testimony about the reliability of eyewitness observations. Whether this is such a case is a debatable question. We must assume on this record that “Vierordt’s Law” is a generally accepted scientific principle; defendant sought, and was denied, a Frye hearing on that issue. The proposition that estimates of the duration of brief incidents tend to err significantly on the high side is not one within the ken of the average juror. And the accuracy of witnesses’ estimates of duration was undoubtedly relevant to this case.
On the other hand, applications to admit evidence of this kind—in essence, testimony by an expert witness advising the jury on how to evaluate the testimony of fact witnesses—must be approached with caution. Such testimony is collateral to the main issues in the case, and we have warned that the exploration of collateral issues tends “to obscure the main issue in the minds of the jury, to lead them away from the principal matters which require their attention and to protract trials to an unreasonable extent without any corresponding advantage to any one concerned” (People v Harris, 209 NY 70, 82 [1913]). This explains the limitations we placed on our holding in LeGrand, requiring the admission of testimony about eyewitness identifications only where uncorroborated identification evidence is of critical importance. Courts do not normally exclude relevant evidence merely because the case against the defendant is strong. But the overall strength of the case is important to issues arising under LeGrand, because where the eyewitness testimony is not crucial, expert testimony about the collateral issue of eyewitness reliability can be a harmful distraction.
Here, it can be argued that it was not crucial for the jury to decide how many seconds or minutes defendant held Reister in a headlock. The evidence that Reister fell unconscious, that defendant still maintained a grip on his neck, and that onlookers screamed for defendant to stop and tried to pull him away without result, would support a finding that, however long it was, it was far too long. The decisive issue in the case is not the *380duration of the headlock, but whether defendant caused Reister’s death while intending to cause him serious physical injury. The theory that defendant’s purpose was only to defend himself might be rejected by a factfinder even in the absence of any evidence of duration.
Yet the People chose to put in much evidence of duration, from fact and expert witnesses, and relied on it heavily. This might well have justified the trial court in allowing Penrod to give the testimony defendant proffered. Whether it was an abuse of discretion to exclude that testimony is now—since we reverse the conviction on other grounds—an academic question. A similar question may arise on retrial, but because no two trials are ever identical the considerations governing the court’s exercise of its discretion will not necessarily be the same. We decide only that the question should be addressed in light of the factors discussed in this opinion.
Accordingly, the order of the Appellate Division should be reversed and a new trial ordered.
Chief Judge Lippman and Judges Graffeo, Read, Pigott, Rivera and Abdus-Salaam concur.Order reversed and a new trial ordered.