Using clinical judgment alone, mental health professionals cannot predict individual patient violence much more accurately than chance. Clinicians could improve their prediction of violence if they routinely used structured risk assessment instruments, but they don't; the use of such tools for screening is not currently the standard of care in the United States and is not commonly reimbursed by insurance. The author argues, however, that clinicians actually can predict and prevent violence if they consider their patients as a group from the perspective of public-health epidemiology. Optimizing treatment for all patients will help prevent violence by the few who pose a risk of violence, even when such patients are not identified in advance.