The prediction of violence in offender populations has long been a dream of correctional decision-makers. The possibility of identifying those individuals who will engage in violent behavior in the future offers the prospect of treatment interventions to reduce such violence. Using elaborate case histories, current measures of mental and emotional functioning, and professional prognoses for a sample of 4,146 California Youth Authority wards, the present study sought to develop a classification device for estimating assaultive potential with sufficient accuracy to be useful in correctional program decisions. Simple classification procedures and multivariate approaches failed to yield an oper ationally practical prediction instrument that would warrant implementation in actual preventive or correctional practice. Much of the violent behavior we would wish to predict will probably never come to our attention and the part that does will be far from a representative sample. The prediction equations themselves contain the seed of self-fulfilling prophecy.