A study of 4,578 appeals during those years showed that most cases ``are so seriously flawed that they have to be done over again,'' said Columbia University law professor James Liebman, the lead author.
``It's not one case, it's thousands of cases. It's not one state, it's almost all of the states,'' Liebman said in an interview. ``You're creating a very high risk that some errors are going to get through the process.''
The study comes at a time of increased debate over capital punishment.
Earlier this year, Republican Gov. George Ryan of Illinois imposed a moratorium on capital punishment in his state after 13 death row inmates were exonerated. Texas Gov. George W. Bush recently approved his first 30-day reprieve in a death penalty case--to allow time for DNA testing--after permitting 131 executions.
But public support for capital punishment remains high. A Gallup Poll in February showed 66 percent back the use of death sentences, down somewhat from Gallup polls during the 1990s that showed support ranging from 71 percent to 80 percent.
The Columbia study said only 5 percent of the 5,760 death sentences imposed from 1973 through 1995 were carried out.
The study, written with professor Jeffrey Fagan and graduate student Valerie West, examined 4,578 death penalty cases in which at least one round of appeals was completed. Of those cases, a state or federal court threw out the conviction or death sentence in 68 percent of the cases.
``Our 23 years worth of findings reveal a capital punishment system collapsing under the weight of its own mistakes,'' the study said.
Richard Dieter of the Death Penalty Information Center, a group critical of how capital punishment is administered, said, ``It's amazing how many mistakes are being made...Those supporting the death penalty might look at it and say this isn't getting us anywhere.''
Dudley Sharp of Justice for All, a Houston-based victims' rights organization, said, ``We all know that these cases get the closest scrutiny imaginable...All systems can be improved and the death penalty is certainly one of those systems.'' But, he added, there has been no proof of an innocent person being executed in the past century.
The study said the rates of reversals varied widely from state to state and among federal appellate circuits.
Texas, which executed 104 people during the study period, showed 52 percent of its death penalty cases reversed on appeal. Florida, which executed 36 people, had a 73 percent reversal rate.
On the other hand, Virginia, which executed 29 people, had a reversal rate of only 18 percent. The study suggested that may be partly due to Virginia's strict limits on appeals in death penalty cases and the overall low reversal rate in the 4th U.S. Circuit Court of Appeals, which hears appeals from that state.
Liebman said the 68 percent reversal rate appears to have come down some since 1995. A second study to be released later this year is expected to look at those numbers, as well as the reasons why death penalty convictions are thrown out.
The main reasons appear to be incompetent defense lawyering and misconduct by prosecutors, the Columbia study said.
Capital punishment resumed in 1977 after a Supreme Court-imposed moratorium, and 313 people were executed by the end of 1995. In recent years, the Supreme Court and Congress have acted to speed up death penalty reviews in federal courts, and there have been 642 executions to date.