For the study reported here, the authors used growth mixture modeling to analyze changes in alcohol and marijuana use and the use of drug and alcohol treatment services for a sample of 180 adolescents with serious emotional disturbance (ages 12-14 years at the beginning of the 7-year longitudinal study). Three latent classes of substance users were identified, two of which exhibited significant linear increases in substance use. As expected, adolescents who were heavily involved in substance use at the beginning of the study and who significantly increased their substance use were the most likely to receive the greatest number of alcohol and drug services. Adolescents who had low levels of substance use at the beginning of the study but whose substance use increased significantly during the study received fewer services. The fact that the initially low substance-use class eventually ended up resembling the high substance-use class at the end of the 7-year study underscores the importance of collecting longitudinal data on substance use and employing the rate of "change in substance use" as an indicator of the need for treatment. Advantages of growth mixture and parallel processing modeling for studying changes in substance use and services are discussed.