Eating disorders are an important and growing health concern, and bulimia nervosa (BN) accounts for the largest fraction of eating disorders. Health consequences of BN are substantial and especially serious given the increasingly compulsive nature of the disorder. However, remarkably little is known about the mechanisms underlying the persistent nature of BN. We use data from a unique panel data set, the National Heart, Lung, and Blood Institute Growth and Health Study, which was conducted for ten years on young women aged 9-10 at the start of the survey (in 1987). Using instrumental variable techniques, we document that unobserved heterogeneity plays a role in the persistence of BN, but, strikingly, up to two-thirds of this persistence is due to true state dependence. Our findings have important implications for public policy since they suggest that the timing of policy is crucial: preventive educational programs should be coupled with more intense (rehabilitation) treatment at the early stages of bingeing and purging behaviors. Our results are robust to different model specifications and identifying assumptions.