Researchers from Michigan State University studied social history data and offense records on all cases referred to the court from 1975 through 1978 -approximately 7,000 referrals for about 4,500 individuals. Six factors confirmed as valid predictors of recidivism by the researchers were type of first offense committed, previous offense record, sex of the offender, quality of home conditions, school behavior and truancy levels, and delinquency by siblings or friends. Multiple regression analysis was performed on the data, supplemented by the development of maximum likelihood estimates concerning the relative impact of each factor. After the relationship to recidivism of each of these six factors was tabulated, their combined relationship was examined to produce a predictive model. Overall, the model correctly predicted 65.1 percent of the study cases. Testing of the model's validity began in 1979. For the next 12, months the actual recidivism of each youngster referred to the court was tracked and measured against the predicted probability of recidivism shown by application of the predictor. The recidivism predictor model demonstrated a high correlation between predicted and actual recidivism. Cost-benefit analysis dictated the allocation of additional resources to the offenders with a high probability of recidivism and a decrease in programs for the low risk group who were successfully predicted not to recidivate 86 percent of the time. The model's utility has been its ability to make a systematic early identification of potential chronic offenders and bring intensive services to them to deter subsequent delinquency. Tabular data are provided.