Longitudinal research on criminal violence has especially investigated the natural history of arrests or convictions for violence among officially processed groups and among more representative samples from the general population. This research shows that convicted offenders are versatile rather than specialized in their crimes and that, while the incidence of violence convictions is low, the prevalence of offenders with at least one conviction for violence may be high. Moreover, the probability of having a subsequent conviction for violence increases with each violence conviction. Longitudinal research has also investigated the prediction of violence, the effects of specific events on it, and its intergenerational transmission. There is continuity between rated aggression at an early age and conviction for violence at a later age. Persons convicted for violence are more likely to have had cold, harsh, disharmonious parents who exercised poor supervision over them; are more likely to have had convicted parents; and are more likely to have had low intelligence quotients. More longitudinal research should be carried out using interviews as well as records with the aim to carry out specifically longitudinal analyses such as those studying the effects of specific events on the course of development. More studies should be concerned with developing theories and testing hypotheses about violence. The most urgent need is to embark on a prospective longitudinal survey of the natural history of criminal violence, with overlapping cross-sectional surveys, and with regular collection of record and interview data. A table and over 100 references are included.