Information used for predictions generally consists of extensive data on attributes, experienced, and activities of individuals. The use of such information for prediction involves two steps. First, a set of individual data is employed to estimate a model. Second, the model is used together with information on individuals and sometimes structural variables to predict future behavior. Prediction models should be estimated using large samples of individuals who are representative of the group for whom predictions are to be made. The most commonly used criterion variable in criminal justice modeling is simply an indicator of whether the individual participated in the activity of interest. The prediction model should incorporate carefully selected explanatory variables and then estimate using parametric survival techniques. The authors strongly oppose the use of currently available statistical predictions to implement a policy of selective incapacitation. 32 references, 2 figures.