Personal variables were divided into three types: drug-use measures, law abidance-related beliefs, and affect-related measures. Social-environmental variables were divided into four types: victimization-related measures, high-risk group identification measures, self-protection measures, and demographics. A baseline violence measure was included as a variable in all prediction models. A total of 29 school districts from a five-county region of southern California were recruited for participation in a previously conducted study, using a procedure approximating random selection. Twenty-one continuation high schools were selected for participation in this study. A total of 870 high-risk juveniles were monitored for violence perpetration by means of self-reports obtained 1 year apart. Self-reported violent behavior focused on the number of times various weapons were used to threaten or injure someone, the number of assaults against persons, and the intentional damaging or stealing of another's property. Factors that predicted later perpetrated violence were marijuana use, a relatively young age, being male, having high-risk group self-identification, low perceived efficacy of the police department, and the nonavoidance of dangerous places. The study thus concludes that personal and social factors beyond baseline violent behavior predict risk for future violent behavior. 3 tables and 37 references