T. E. Moffitt's (1993a) hypothesis that adolescent-limited criminal offenders will have higher scores on tests of cognitive ability than life-course-persistent offenders was tested with 12 tests of cognitive ability given to a large and diverse sample of delinquent juveniles whose arrest records were collected over 20 years. This is the first investigation to empirically evaluate this proposal with longitudinal data obtained from a sample for a long enough time to distinguish life course patterns of crime. This study provided only partial support for Moffitt's hypothesis because the results varied by ethnicity. We found relatively consistent support for the hypothesis for Caucasians and Hispanics but no support for the hypothesis for African Americans. These findings are interpreted in terms of differences in developmental contexts for individual ethnic groups.