The results of the two-part growth model indicate that youths' frequency of marijuana use was maintained over the mid to late adolescent years. The significant positive relationship between the intercept growth factors for part 1 and part 2 of the model points to a latent tendency to engage in use and a conditional, continuing decision on frequency of use. Youths who are less likely to initiate marijuana use were also less likely to use the drug frequently. Failure to model this phenomenon in longitudinal studies can result in substantial bias in coefficient estimation. Findings underscore the importance for researchers conducting longitudinal research on drug use to consider including the estimation of two-part growth models in their analyses. This is particularly the case for intervention studies; findings did not indicate a significant difference in the probability of and frequency of marijuana use trajectories when comparing delinquents who participated in the Youth Support Project (YSP) intervention with those in the control group (treatment as usual from the juvenile justice system). Data were collected from 278 youths arrested and processed at the Hillsborough County Juvenile Assessment Center (JAC) who agreed to participate in the YSP. Tables, figure, and references