Follow-up studies of the physical dangerousness of men released from correctional institutions, mental hospitals and special security psychiatric institutions have indicated that we cannot as yet predict with an acceptable degree of accuracy which men will be dangerous upon release. Because of the methodological and conceptual weaknesses of these predictive studies and their lack of success, as well as the practical difficulties which would arise from not attempting to select nondangerous men for release, a new approach to the problem of prediction is required. Such an approach might profitably make predictions according to how different sorts of offenders respond to various situations. Theories taking into account the interactions between persons and their environment may in future allow tolerably accurate predictions of future assaultive behavior under certain circumscribed conditions.