This study explores the emergence of a criminal career in adulthood. The main hypothesis tested is that late criminal onset (at age 21 or later) is influenced by early factors that delay antisocial manifestations. The Cambridge Study in Delinquent Development (CSDD) was used to examine early determinants of criminal behavior. 400 Inner London males were followed from ages 8-10 to 48-50, and were classified as follows: 35 late onsetters who were first convicted at age 21 or later, and did not have high self-reported delinquency at ages 10-14 and 15-18; 129 early onsetters first convicted between ages 10 and 20; and 236 unconvicted males. Odds ratios and logistic regression analyses revealed that the best predictors of late onset offenders compared with early onset offenders included nervousness, having few friends at ages 8-10, and not having sexual intercourse by age 18. The best predictors of late onset offenders compared with nonoffenders included teacher-rated anxiousness at ages 12-14 and high neuroticism at age 16. It is concluded that being nervous and withdrawn protected boys against offending in adolescence but that these protective effects tended to wear off after age 21. These findings show that adult offending can be predicted from childhood, and suggest that early intervention might prevent a variety of maladjustment problems and difficulties in adult life.