Previous research has revealed heterogeneity within the offender population and among offending trajectories. This research has also indicated the importance of considering separate phases of the offending cycle from onset to desistance. This study used data from the Seattle Social Development Project. The study population included all 808 students who consented to participate from among the 1,053 fifth grade students in 18 Seattle public elementary schools. The study overrepresented students from high-crime and low-income neighborhoods. This longitudinal study followed the youths starting in 1985, when most were approximately 10.5 years old, through the spring of each succeeding year through 1991, when most participants were age 16. The analysis used semi-parametric, group-based modeling to identify five offense trajectories: nonoffenders, late onsetters, desisters, escalators, and chronic offenders. The analysis then used multinomial logistic regressions to examine childhood predictors measures at ages 10 to 12 that distinguished these 5 groups. Results indicated that among initial nonoffenders at age 13, individual factors distinguished late onsetters from nonoffenders. In addition, among youths who were already delinquent at age 13, peer, school, and neighborhood factors distinguished escalators from desisters. Findings clearly suggested that prevention efforts should begin early and that early interventions can be the most effective approaches for both preventing the initial onset of offending for children with no prior history and preventing progression to serious offending for children already exhibiting low levels of offending. Findings also suggested the need for multicomponent prevention strategies that span several domains of childhood. Tables, figure, notes, appended list of measures and predictor variables, and 62 references (Author abstract modified)