Statistical prediction differs from clinical prediction both in the kinds of data it uses and in the methods it uses to convert the data into a prediction. Statistical prediction uses lower order, often demographic, variables and combines them by means of automatic mathematical rules. Clinical prediction, on the other hand, is less precise about the predictor variables used and may choose different predictors for different cases. These factors are then transformed into a prediction in a subjective or intuitive way. The factors most closely related to the occurrence of violent behavior are past violence, age, sex, race, socioeconomic status, and opiate or alcohol abuse. Estimated IQ, residential mobility, and marital status also are related to violent behavior. Mental illness, however, is not apparently related to violence in the absence of a history of violent behavior. Clinicians who wish to improve the accuracy of their predictions by incorporating statistical information can best do so by making the base rates of violent behavior a prime consideration. 2 tables, 64 references.