Previous research has shown heterogeneity in offense trajectories. Using data from the Seattle Social Development Project, a longitudinal study of 808 youths followed since 1985, this study seeks to identify childhood predictors of different offense trajectories. Five offense trajectories were identified using semiparametric, group-based modeling: nonoffenders, late onsetters, desisters, escalators, and chronic offenders. Multinomial logistic regressions were then employed to examine childhood predictors measured at ages 10 to 12 that distinguish these five groups. Results indicated that among initial nonoffenders at age 13, late onsetters were distinguished from nonoffenders by individual factors. Among youth already delinquent at age 13, escalators were distinguished from desisters by peer, school, and neighborhood factors.