Compliance with drug therapy is of major concern to clinicians as well as policy makers since uncontrolled symptoms due to noncompliance present health risks for patients and may lead to social costs. Noncompliance comes in the form of skipped dosages as well as discontinuation well before a clinician deems it appropriate. The problem is especially severe in behavioral disorders among children where the symptoms can last well beyond adolescence. We use pharmacy dispensing and clinical diagnosis data on children diagnosed with attention-deficit hyperactivity disorder (ADHD) and who are on ADHD-related medications. The paper shows how the pharmacy refill data fit naturally into a discrete time hazard rate framework, and then compares estimates from alternative definitions of discontinuation. We use a long follow-up period (up to 6 years), allow for a flexible duration dependence and account for unobserved heterogeneity. The expected duration is about 18 months with significant differences across race, gender, copays, medication switching, and seasonality. We find that African-American, Hispanic and, Asian children are about 39% more likely, on average, to quit therapy in a given month than white children. Similarly, compared to a child that initiates drug therapy at age 9, a child that starts therapy at age 10 is 26.4% more likely to discontinue at any given time. Earlier literature using the hazard approach reports smaller associations between these covariates and durations. We show that this could be because of ignored unobserved heterogeneity, use of a relatively short follow-up study design and monotonic duration dependence. Finally, our results are of particular relevance to clinicians as well as to policy makers given recent changes in federal and state policies that may make early detection and diagnosis of ADHD among children less likely.