The study of the Philadelphia bail project and a number of sentencing guideline studies are reviewed. The review indicates that decisionmakers tend to rely on a few common items of information regardless of the decision being made. Descriptive (based on past decisions) and normative (based on selected criteria) studies apparently recommend different items of information as predictive of offender behavioral outcomes. The evidence indicates that criminal justice decisionmakers do not make optimal decisions in predicting offender behaviors. Still, a number of factors have some predictive validity across a variety of settings, and decisionmakers do not focus on these factors. Despite substantial base rate problems, most guidelines have yielded prediction that exceeds the chance rate. The evaluations of guidelines effectiveness often ignore the dependence of guidelines effectiveness on proper development and implementation. Fundamental conflicts between normative and descriptive models, however, may constrain effectiveness in many applications of guidelines. 5 notes and 42-item bibliography. (Author abstract modified)