OBJECTIVE: The authors evaluated the reliability and validity of a probabilistic approach to clinical assessment of short-term risk of violence. METHOD: At admission, nurses and physicians independently rated the probability that each of 149 psychiatric patients would physically attack someone during the first week of hospitalization on a university-based locked inpatient unit. Ward behavior was measured with the Overt Aggression Scale. RESULTS: There was a moderate level of agreement between nurses' and physicians' assessments of risk. Ratings of ward behavior showed an increase in the proportion of assaultive patients as the level of estimated risk of violence increased. Although the overall rate of assaults was overpredicted, there was a close correspondence between clinical estimates of patients' chances of becoming violent and the proportion of patients within each risk level who later displayed some type of inpatient aggression. CONCLUSIONS: The reliability and validity of short-term estimates of the risk of violence among acutely disturbed inpatients may be higher than past violence research has suggested. These findings provide preliminary support for the utility of a probabilistic approach to assessment of the risk of violence.