This study tested the predictive validity of a novel, brief, and easy-to-use self-report measure of expectancies and their subjective values for alcohol and marijuana use. Canadian students in Grades 7 to 11 were administered paper-and-pencil questionnaires once per year for 3 consecutive years (Krank et al., 2011). As part of the questionnaire, participants completed an outcome expectancy measure where they were asked to list 3 or 4 things they expected would happen if they used a particular substance (i.e., alcohol, marijuana) and to indicate for each whether they would or would not like this outcome. “Liking” outcomes were coded as +1, “not like” as −1, and summed to obtain an outcome expectancy liking (OEL) sum for each participant and each substance. Participants also completed substance use behavior questions for alcohol and marijuana. Multilevel modeling demonstrated that OEL sum significantly predicted the intercept and slope of substance use trajectories by participants, even when demographic variables were controlled. For both alcohol and marijuana, multilevel modeling analyses indicated that a more positive OEL sum for a substance in the first year of the study were more likely to have tried that substance earlier (intercept) and were more likely to escalate their use at a greater rate over time (slope). The results complement the predictive validity found with other direct and indirect measures of substance use associations. The outcome expectancy liking task is a simple and unobtrusive method for identifying adolescents who are at risk for early substance abuse.