In both criminal and civil justice systems legislatures and courts have produced statutes and case law that endorse the validity and appropriateness of experts' prediction of violent behavior. Behavioral scientists are expected to make predictions about human behavior to validate their professional expertise, and the predictions do not have to be accurate; they must only be believable. There has been no empirical demonstration by any profession, using a clinical or actuarial approach, that the accuracy of predicting violence in the criminal justice or mental health systems is better than that attainable by chance. More accurate predictions of violence are difficult because of the relatively low frequency of violent behavior and the inaccuracy and incompleteness of data used as predictors. Although for certain potential victim classes, such as public figures, offender mental disorders may be significantly related to violent behavior, this is not a significant factor in predicting most violent behavior. Any effort to predict violent behavior should consider contextual variables in addition to individual characteristics. This requires the use of a multiplicity of variables and experts to render a prediction. Also, a prediction should present empirical facts about the probability of the prediction being accurate. 29 notes and a 'roundtable' response to the paper.