Abstract: Several recent and ongoing large-scale studies of child and adolescent development have succeeded in collecting rich longitudinal data from psychosocial, behavioral and neural levels of analysis. Although these projects offer researchers an unprecedented opportunity to investigate developmental factors that contribute to mental health outcomes, they also present significant challenges due to the need to draw interpretable conclusions from high-dimensional data and integrate measurements over several levels of analysis. The emerging field of computational psychiatry, which emphasizes the use of mathematically-specified models for measuring clinically-relevant mechanistic processes that underlie observed behavioral and/or neural data, offers potential solutions. I will present a brief overview of this approach and two specific applications. The first involves the use of a mathematical model of go/no-go task performance to clarify mechanistic processes indexed by task-related neural activations in late adolescence and inform the prediction of substance use in emerging adulthood. The second involves efforts to use linear growth modeling to assess relationships between pubertal timing, risk for substance abuse in adolescence, and individual differences in reward evaluation circuitry which may mediate that risk. These lines of research suggest that quantitative model-based approaches can facilitate the use of large-scale longitudinal data sets to better understand and predict externalizing psychopathology.