Within a 4-month time frame, 242 patients were asessed as to their potential for dangerousness using ratings of a clinical assessment team. Fifty-four of the patients were remainded for further inpatient analysis requiring anywhere from 4 to 52 days. Since the behavior of inpatients is recorded each 8-hour shift in log format, the possibility of conducting some preliminary testing of dangerous predictions presented itself. For study purposes, an incident of dangerousness was defined as any aggressive act which involved physical contact with another person whether or not injury resulted, as well as threatening gestures without physical contact and verbal threats against specific persons. Of the 52 patients in the sample, 23 were responsible for some kind of dangerous incident, while 29 remained on the inpatient unit without incident. On first inspection, the prediction data collected from the two external coders involved in the original assessment appears to show impressive validity in predicting who will and will not manifest future dangerousness. The presence of a large number of false positive predictions (unfulfilled predictions of dangerousness), however, detracts substantially from prediction accuracy. The study's limitations are indicated. Tabular data and three references are provided.