This article examines the well-documented relationship between early initiation or onset of criminal behavior and a heightened risk of involvement in offending. Previous research examining this question conducted by Nagin and Farrington ("Criminology" 30:235-260, 1992a; "Criminology" 30:501-523, 1992b) used data from the Cambridge Study in Delinquent Development and found that: (1) onset age was correlated with offending involvement; and (2) the correlation could be explained by stable individual differences in the propensity to offend rather than a causal effect of early onset age. In this study, similar analytic methods are applied to data from the Second Philadelphia Birth Cohort. This data set consists of all 13,160 males born in Philadelphia in 1958 who resided in the city continuously from ages 10 to 18, slightly more than half of whom were non-white. Information from each of the youths was collected from schools, juvenile justice agencies, other official sources and surveys. In a model that mimics previous analyses, we initially found that an early age of onset is associated with greater subsequent involvement in delinquent behavior. When unobserved criminal propensity was controlled, however, we found that a late rather than an early onset of delinquency was related to future offending. In finding a state dependent effect for age of onset, our findings are contrary to propensity theory in criminology. In finding that it is late rather early onset which puts youth at risk for future offending, our findings are contrary to developmental/life course theory. Our results are more compatible with traditional criminological theory that is friendly to state dependence processes, though they too have not to date articulated why a late onsetting of offending might be particularly criminogenic.