This study compared the accuracy of an actuarial procedure for the prediction of community violence by patients with mental illness with the accuracy of clinicians' ratings of concern about patients' violence. Data came from a study in which patients were followed in the community for 6 months after having been seen in a psychiatric emergency room. Accuracy of actuarial prediction was estimated retrospectively, with a statistical correction for capitalization on chance. Actuarial prediction had lower rates of false-positive and false-negative errors than clinical prediction. The seriousness of the violence correctly identified by the actuarial predictor (the true positives) was similar to the seriousness identified by clinicians. Actuarial predictions based only on patients' histories of violence were more accurate than clinical predictions, as were actuarial predictions that did not use information about histories.