Regarding the stability of co-offending, the sample is best characterized by multiple co-offending trajectories. All models presented in the analysis showed some high-level and low-level co-offending pathways over the juvenile's criminal career, as well as classes that were consistent with the general pattern of individuals showing a reduced tendency to co-offend with age; however, this latter pattern did not hold true for all classes of youth. Aggregate indicators of co-offending over time were found to be insensitive and misleading analytically, because they masked unique and varying developmental trajectories. The analysis of the stability of selecting co-offenders also found that aggregate analyses were problematic. Although the data showed that most offenders tended not to reuse the same co-offenders in crimes, there were variations in this pattern. Subjects with more arrests and who offended with larger groups were more likely to reuse particular individuals as co-offenders. Among those subjects who did reselect co-offenders, more frequent offenders and those who tended to offend in larger groups showed lower levels of co-offender stability over their juvenile criminal careers. The primary sample of 400 subjects was predominately male and African-American, and nearly 84 percent of the sample had at least one co-offense. The "stability of co-offending" was measured by the number of co-offenses in which the subject was involved for every year from ages 10 through 17. The stability of co-offenders was measured as the tendency of each subject to reuse co-offenders. 2 tables, 2 figures, 78 references, and appended descriptive statistics for the sample