The prevalence of violence can be relatively low or relatively high, depending on how it is measured (e.g., by convictions or self-reports) and on the length of the time period during which it is measured. Also, the accuracy of prediction of violence can be relatively low or relatively high, depending on how predictive efficiency is measured (e.g., focusing on false positives versus odds ratios) and on whether predictions are viewed as deterministic or probabilistic. Similar points can be made about the prediction of non-violent offending. The article illustrates these points using data collected in a prospective longitudinal survey of the development of offending and antisocial behavior in 411 London males aged 8-9 years. Identification of a high-risk category of people can be based on either antisocial behavior or a deprived background or both; the odds ratio is a better measure of predictive efficiency. Tables, references