HUNDREDS of people watched the crash of American Airlines Flight 587 near Kennedy International Airport in New York on Nov. 12, and in the course of 93 seconds they apparently saw hundreds of different things.

According to the National Transportation Safety Board, which announced this month that it had gathered 349 eyewitness accounts through interviews or written statements, 52 percent said they saw a fire while the plane was in the air. The largest number (22 percent) said the fire was in the fuselage, but a majority cited other locations, including the left engine, the right engine, the left wing, the right wing or an unspecified engine or wing.

Nearly one of five witnesses said they saw the plane make a right turn; an equal number said it was a left turn. Nearly 60 percent said they saw something fall off the plane; of these, 13 percent said it was a wing. (In fact, it was the vertical portion of the tail.)

The investigators say there is no evidence in the wreckage or on the flight recorders of an in-flight fire or explosion. A plane breaking up in flight, as this one did, might in its last moments produce flashes of fire from engines ripping loose, but the idea that the plane caught fire is a trick of memory, they say.

None of this is surprising, said Dr. Charles R. Honts, a professor of psychology at Boise State University and the editor of the Journal of Credibility Assessment and Witness Psychology. "Eyewitness memory is reconstructive," said Dr. Honts, who is not associated with the safety board. "The biggest mistake you can make is to think about a memory like it's a videotape; there's not a permanent record there."

The problem, he said, is that witnesses instinctively try to match events with their past experiences: "How many plane crashes have you witnessed in real life? Probably none. But in the movies? A lot. In the movies, there's always smoke and there's always fire."

As a result, the safety board generally doesn't place much value on eyewitness reports if data and voice recorders are available. For many investigators, the only infallible witness is a twisted piece of metal.

Benjamin A. Berman, a former chief of major aviation investigations at the safety board, said pilots actually make the worst witnesses, because their technical knowledge can lead them too quickly to identify a mechanical problem that may not have occurred. "Children make among the best witnesses," he added, "because they don't tend to place an interpretation on what they've seen."

The safety board's skepticism of eyewitness accounts was deepened by the explosion of TWA Flight 800 off Long Island six years ago: hundreds of people saw an upward streak that they assumed was a missile, although investigators said it was the body of the plane itself, streaking upward after the forward portion had fallen off following a fuel tank explosion.

THAT disaster highlighted another pitfall for investigators, Mr. Berman and others say: F.B.I. agents asked witnesses where the missile came from, presupposing the presence of a weapon. "It wasn't good aircraft accident investigation," Mr. Berman said.

There are other well-known cases of witness error, including the crash of a Lauda Air Boeing 767 near Bangkok in May 1991. Witnesses said they heard a bomb and saw the plane fall in flames, but it turned out to be a mechanical problem.

So why do investigators bother asking witnesses at all? Dr. Bernard S. Loeb, who retired as the safety board's director of aviation safety last year, said, "In the case of 587, it's unlikely that the witnesses will provide much to help the investigation, but you never know that when you begin an investigation — where you're going to get important leads, from the recorders, from witnesses, from the structure itself."

And in any crash, he said, conflicting witness statements can still be useful. "What was very clear from the Flight 800 witnesses was that many did see something up in the sky," he said.

Even if the accounts are likely to be wrong, they are still routinely gathered and evaluated by both the board and police agencies. "Can you imagine if we didn't interview the witnesses?" said one current board official.

Mr. Berman, who left the board last year, said investigators may have released the summary of what the Flight 587 witnesses saw just to show publicly that the statements showed "scatter" — an engineering term for plotted data that does not fit a pattern. A release at this late date is unusual, but a spokesman for the board, Ted Lopatkiewicz, said it was done because it was ready. But, he added, "I don't think I'm making any news by saying that eyewitness testimony at a plane crash and probably at many traumatic events is unreliable."

Witness statements can be more valuable in crashes of small planes that don't have flight data recorders or cockpit voice recorders, Mr. Berman said.

Mr. Loeb said his experience with witnesses had led him to question the reliability of criminal convictions based on eyewitness identifications. In Illinois, he noted, a commission appointed by the governor recommended in April that the death penalty not be applied to murder convictions based on a single eyewitness identification.

Mr. Loeb said his personal experience also played into his skepticism. Recently he and his wife saw a two-vehicle collision, and unlike plane crash witnesses, they both saw it from the same angle. Within moments, they disagreed about what they had seen. Among other key details, Mr. Loeb said he could not recall whether one of the vehicles had been a truck or an S.U.V.