The sample was recruited when the boys were in grade 4 and consisted of two cohorts to permit replication of constructs. The sample was selected from schools in the higher crime areas of a medium-sized metropolitan region in the Pacific Northwest; they were considered at risk for later delinquency. Cohort I (n=102) was recruited in the 1983-85 school year, and Cohort II (n=104) was recruited in the 1984-85 school year. The two cohorts were similar on all major variables and were combined for the analyses. Data-collection procedures consisted of interviews with parents and subjects, observation of subjects at their homes, questionnaires for the parents and the subjects, the administration of the Wide Range Achievement Test, and peer assessments of subjects. Data on offending were obtained from arrest records and self-reports. Family characteristics, including family-management practices and childhood behavior, were compared for violent adolescent arrestees and nonviolent adolescent arrestees, who were matched for arrest frequency. Regression analyses were conducted by using self-reports of violent offending. The study also hypothesized that multiple arrestees with no arrests for violence would self-report as much violence as multiple arrestees with arrests for violence. The findings generally support the contention that violent offenders have the same background as frequent but nonviolent offenders. This suggests that programs to prevent violent behavior should be similar to programs that prevent chronic offending. 5 tables and 56 references