In this review, the authors examine the research evidence for the prediction of wife assault recidivism, lethal wife assault, and wife assault onset. They also review and present original data on the effect of treatment attendance on wife assault risk. Violence does not always become a stable habit, and variables associated with wife assault onset do not necessarily predict recidivism. General antisociality, psychopathy, substance abuse, and a history of assault and psychological abuse in the relationship are the most promising predictors of recidivism. Formal risk assessments, and victims' predictions, have demonstrated value in predicting recidivism. The authors review existing assessments for wife assault onset and recidivism and explain the relative merits of actuarial tools and structured clinical assessments. Because of statistical and practical limitations to predicting lethal assault, they recommend using an actuarial assessment of wife assault risk, plus attention to the strongest correlates of lethal assault when lethality is a concern.