The article explains how the accuracy of a test or detection system should be measured and what accuracy measurements mean. It also summarizes evidence for the proposition that clinical judgments about future violence have better-than-chance accuracy. It examines the practical import of available violence prediction methods, i.e., what those methods tell clinicians about the likelihood of violence in populations that they evaluate or treat. The discussion reveals what seems to be a paradox: even reasonably accurate assessment instruments may not have much practical value for clinicians who make decisions about violent patients. The article explores that finding and its implications for clinical decision making. Currently available violence prediction techniques are not accurate enough to sort patients into subgroups with meaningfully different levels of risk. However, mental health professionals can help reduce violence by encouraging outpatient commitment and close community follow-up and by educating families and intensively following patients after discharge. Notes, figures, tables, references