Psychiatrists and psychologists have available structured risk assessment instruments to assess the risk of patient violence. These instruments are also used to help make important legal decisions, including which prisoners will be evaluated for continued detention at the end of their sentence. The predictive validity of structured instruments has been demonstrated in operationally defined groups. Their application to individual cases has led to objections that the standard deviations for the risk categories generated by the instruments overlap significantly. This debate has paid insufficient attention to the differences between aleatory (statistical) and epistemic (degree of confirmation) approaches to uncertainty. The approach to uncertainty in psychiatric violence risk assessment is, of necessity, largely epistemic. Providing statistical data can only be part of establishing the precision of an estimate of the probability of someone acting violently. Copyright © 2012 John Wiley & Sons, Ltd.