The contribution of selected neurological, psychiatric, and demographic variables to the frequency of violent incidents was studied in an institutional context. A multiple regression procedure was used to predict the number of documented incidents for each of 45 neuropsychiatric patients. This procedure yielded a 5-variable equation that accounted for close to a third of the variance. The first three variables entered each contributed uniquely to validity. These variables were presence of focal frontal cerebral lesions, number of inpatient days, and history of seizure disorder. The presence of cerebral damage in general did not predict violent incidents. The results are discussed in terms of possible brain-behavior relationships and the nature of the patient population.