This study investigates several factors as possible mediators of physical child abuse in the prediction of violence among adolescents. Prospective and retrospective measures of abuse are compared in mediation tests. Data are from the Lehigh Longitudinal Study, a prospective study of 457 children followed from preschool into adolescence. Structural equation models examined the degree to which abuse is mediated in the prediction of violence through youths' bonds to family, commitment to school, involvement with antisocial peers during adolescence, and attitudes about the use of violence. The model included measures of family socioeconomic status and youths' gender and age as controls on violence. Findings suggest that abuse (whether measured prospectively or retrospectively) is heavily mediated in its prediction of later violence and that a sizeable proportion of variance is accounted for in the violence outcome. A fuller pattern of mediation was shown when the retrospective abuse variable was modeled.