The importance of classification in criminal justice research and policy is demonstrated in a historical review of classification methods. The paper examines rationales underlying the use of prediction in criminal justice systems, the validity of prediction research, and prediction as a management tool. Two applications of classification and prediction are discussed: (1) in research, evaluation, and planning and (2) in individual decisions. The paper assesses the following strategies for improving predictive classification validities: fitting methods to assumptions, measurement and scaling, outcome criteria, dealing with law base rates, and developing and applying theory. Other recommendations address extending the use of statistical design, comparative studies of classification methods, developing new variables, measuring stakes as well as risk, and reducing errors in decisions. Over 60 references.