The first of three studies analyzed by the Special Projects Research Unit (SPRU) in order to evaluate clinical or psychiatric predictions of dangerous behavior was the group of 967 dangerous, mentally ill patients transferred from maximum security to regular security prisons as a result of Baxstrom v. Herold and monitored over a subsequent 4-year period. One limitation of the study was that there was no explicit psychiatric prediction of dangerousness. Findings from a second 3-year study, in which determinations of dangerousness were made for 257 felony defendants by 2 psychiatrists at the direction of a New York criminal court, were more a reflection of the high occurrence of assault rather than of any increased predictive ability. A third study conducted at the Patuxent Institute for Defective Delinquents resulted in nearly identical second arrest rates between groups evaluated as dangerous and as not dangerous. The use of statistical prediction on the data of the first two studies resulted in improved predictions of dangerousness. A review of other studies of dangerousness to integrate other empirical findings with those of the studies being evaluated shows that psychiatrists have not demonstrated any special ability to predict dangerous criminal behavior. Also, whether the findings of all studies are clinically or statistically derived, dangerous behavior is vastly overpredicted. However, statistical predictions are consistently superior to clinical predictions gauging relative risks to the individual and to society. In conclusion, community situations to which the released inmate is returning should be taken into account in predictions of dangerousness; and the psychiatrist remains an important source in the development of more accurate prediction devices. Eighteen references and one footnote are provided.