This study presents and analyzes data on the first 873 exonerations reported by the National Registry of Exonerations, exonerationsregistry.org, a new project of the University of Michigan Law School and the Center on Wrongful Convictions at Northwestern University School of Law. The Registry assembles and posts information on exonerations of people who were convicted of serious crimes in the United States. The database, which includes exonerations from January 1989 through February 2012, is larger and more diverse than any comparable collection. Over 60% of the cases did not involve DNA evidence. Earlier datasets were limited almost entirely to rape and murder cases, but more than 150 of these exonerations were for convictions that did not involve homicide or sexual assault, and more than 100 of the sexual assault exonerations were from child sex abuse convictions. The study also describes over 1100 additional 'group exonerations' that occurred in the aftermath of 12 scandals around the country in which police officers systematically planted or fabricated evidence to frame defendants for non-existent crimes, usually drug crimes. The major findings of the study include: (1) The composition and distribution of the exonerations we know about lead to the inescapable conclusion that they are a small fraction of all false convictions for serious felonies. (Exonerations for lesser felonies and misdemeanors are almost entirely absent from the data.) The known cases also strongly suggest that most exonerations escape notice as well. (2) The causes of false convictions vary greatly by crime. For homicides, which account for nearly half of all exonerations in the study, the leading cause of error was perjury and other false accusations usually deliberate false identification of the defendant as the criminal. Homicide exonerations also include a high rate of official misconduct and three-quarters of all false confessions in the study. For adult sexual assault cases, and for the much smaller number of robbery exonerations, the leading cause of error was mistaken witness identifications. Sexual assault exonerations also include a large number of cases with false or misleading forensic evidence. Most child sex abuse exonerations, by contrast, were based on fabricated crimes: the defendants were accused and convicted of crimes that never occurred. Judging from these data, the conviction of innocent defendants is not a single problem but several distinct problems, depending on the nature of the underlying crime.