A model designed to predict early arrest is presented. Family process variables were hypothesized to be indirectly related; a childhood measure of an antisocial trait was assumed to be directly related to risk for early onset. Longitudinal data were collected for two cohorts of families living in high-crime areas of a medium-sized metroplolitan area. Multi-agent/multimethod definitions for both the family process and the antisocial trait concepts defined the latent constructs. Official records of police contacts were collected during the ensuing 5 years. The data for the hazard rates showed that risk for police arrest varied significantly as a function of child’s age. A continuous time regression analysis showed that most of the individual family process constructs and the antisocial trait construct were significantly related to an altered risk for police contacts. A multiple regression analysis showed a similar effect in predicting age at onset. In keeping with the hypotheses in both analyses, when all variables were entered simultaneously the contributions of the family process constructs were nonsignificant; only the contributions for the trait score and social disadvantage remained significant. The more antisocial the child, the greater the risk for early arrest.