Thornberry's interactional model (1987) suggested that the process of delinquency could be explained by the reciprocal relations between social control variables and social learning variables over developmental stages. However, previous studies for testing interactional model (Thornberry et al., 1991, 1994) had some limitations. They did not adequately include theoretically significant measures of both social control and social learning variables. They did not fully cover the transition throughout adolescence. They also did not adequately examine the variation of social category and individual criminal propensity. This study selects family attachment as social control and deviant peer association as social learning variable. This study tests the reciprocal relationships among family attachment, delinquent peers, and delinquency across the full adolescence and early adulthood. In addition, this study hypothesizes that causal processes vary by gender and the early and late onset group. Data to test for these hypotheses are from 1977, 1980, 1983, and 1987 of the National Youth Survey. This study found that deviant peer association is more reciprocally related to delinquency than family attachment. The findings also suggest that gender and criminal propensity are important to determine deviant behavior processes. The effect of family attachment on delinquency is stronger for females than males. The effect of deviant peers on delinquency is stronger for males than females. Female delinquency has stronger negative effect on family relations than male delinquency. The early onset group has more stable levels of delinquency than the late onset group. The delinquency of the early onset group is primarily influenced by family attachment and deviant peers and their relationships are reciprocal, whereas the delinquency of late starters is mainly influenced by deviant peers and their relationships are unidirectional. This study suggests that delinquent processes vary by age, gender and different types of offending.