In a Tampa courtroom in 1981, Alan Crotzer watched as five people took the stand and swore that he was one of the men who kidnapped a 12-year-old girl and 38-year-old woman during a botched home invasion. The child and woman testified that he later raped them.

But all five were wrong, and after spending 24 years in prison, Crotzer was set free in 2006 when DNA testing proved he couldn't have been the rapist.

"Five people were 100 percent convinced they were correct, and each was 100 percent wrong," said Sam Roberts, an assistant public defender in New York City who would later help free Crotzer.

Crotzer's case is typical of 75 percent of the DNA exonerations in Florida and nationwide: Innocent people were sent to prison, in part, because eyewitnesses got it wrong.

That's true in nine of Florida's 12 cases, according to the Innocence Project of Florida, a nonprofit that works to free the wrongfully convicted. It's also true of 198 of the 261 cases nationally, according to Gary Wells, an Iowa State University professor of psychology who has studied witness misidentification for more than 30 years.

How can so many people be so wrong, especially when so much is at stake? The Florida Innocence Commission, a panel appointed by the Florida Supreme Court, will hear evidence about that Nov. 22 in Tampa. Its job is to study those 12 wrongful convictions and come up with ways to prevent more.

One thing already clear about those Florida mistaken eyewitness cases: The problem usually began when police showed witnesses a photo or photo lineup, which is what happened in the Crotzer case.

"Memory is very fragile," said Wells. "It's delicate. It can deteriorate. It can be contaminated."

Best practices can help avoid mistakes

Psychologists, researchers and the Department of Justice have each come up with recommendations on how best to conduct photo and live police lineups to minimize witness mistakes.

One of the keys, according to psychologists and researchers: Have someone who does not know which photo is the suspect show the lineup.

Another, psychologists say: Don't show all the photos at once. Display one, then remove it and show the next.

Florida law-enforcement agencies do not, as a rule, follow those recommendations, although the sheriff's offices in Broward and Hillsborough counties do. So do the Fort Lauderdale police.

"There is no standardized protocol that's been adopted by the law-enforcement agencies statewide," said Lester Garringer Jr., executive director of Florida's Innocence Commission.

New Jersey and North Carolina have adopted those best practices. So have law-enforcement agencies in Boston, Dallas, Denver, Minneapolis and Santa Clara County Calif., said Wells, the Iowa State expert.

He has conducted hundreds of experiments, staging a crime, then asking witnesses to identify the perpetrators. His findings jibe with those of several other studies on live police lineups: Witnesses get suspect IDs wrong about 20 percent of the time, he said.

There are logistical reasons witnesses make errors: It may have been dark. They may have only seen the suspect for a few seconds. He may have been a good distance away.

And there are cognitive reasons: There may have been a weapon on which they were focused.

Also, people have trouble recognizing faces they have only seen once, especially if it was during a time of great stress or fear and especially if that person is of another race, according to psychological research.