A considerable theoretical, empirical, and clinical literature has accumulated on the prediction of dangerous and violent behavior. Despite the pleas of mental health professionals that the prediction of violent behavior is not clinically feasible, courts upholding a therapist's duty to protect third parties have created a legal duty to do so. The first generation of research studies of the clinical prediction of violent behavior focused on long-term predictions in offender populations, of little direct relevance to the setting of outpatient therapy with nonoffenders. More recent clinical research has centered on predictions of imminent violence in noncriminal populations, particularly in the process of involuntary hospitalization, but this reveals relatively low levels of predictive accuracy. Methodologic and ethical factors also limit the generalizability of such work. Given the likelihood that significant numbers of false positive predictions will result, especially in the low base rate population of outpatient treatment, it is suggested that therapists attempt to minimize the antitherapeutic influence of any breach of confidentiality.