Intense debate has focused on the use of statistical predictions of dangerousness in the criminal justice system. Two conflicting positions maintain wide support: that such predictions are never appropriate in criminal justice decision-making, and that they should be used far more often. Recognizing the fact that implicit and intuitive predictions are made every day in police, prosecutorial, sentencing, and other decisions, and explicit but unscientific predictions are common, this article suggests a theoretical framework justifying limited use of statistical predictions. Statistical predictions may present, in some instances, a morally preferable alternative to biased nonscientific and implicit judgments. Development of a sound jurisprudence of predictions faces major hurdles given the trend toward unscientific predictions in the law and the enormous judicial confusion in dealing with predictions. The concept has contributed to a string of notably poor Supreme Court decisions.