Probation departments across the United States, Canada, and Europe use assessment instruments to predict who is most likely to engage in criminal behavior. This dissertation suggests that prediction and classification methodologies need constant improving in order to be useful. An overview of prediction and classification techniques is presented, and the methods used in this study, the construction of several risk prediction models to compete against the Wisconsin risk and need assessment, the most widely used risk assessment in the field of community supervision are described. The dissertation provides a description of the various types of offenders placed on probation in Texas in 1993, and explains that the three potential dependent variables used in developing effective prediction models are successful probation, re-arrest, and probation revocation. Testing the validity and reliability of predictor variables, the author indicates that four index scales, education, employment, substance abuse, and criminal history, are the most effectively used predictors of recidivism. In terms of constructing re-arrest, revocation, and successful probation prediction models, after documenting the poor performance of the Wisconsin assessment in predicting future criminal behavior, the author suggests that using a battery of assessment grounded in theory is the most effective technique for predicting and classifying the individuals qualified for community supervision. The author contends that this approach takes into account the offender’s characteristics and the community in which he or she resides, more effectively predicting offenders’ likelihood of re-offending. The author suggests that significant improvements can be made to offender risk prediction instruments if such instruments are linked to criminology theory. An extensive series of appendices presenting the felony cohort data, codebooks, and data results used in this study completes this dissertation.