Medicare Part D enrollees face a complicated decision problem: they must dynamically choose prescription drug consumption in each period given difficult- to-find prices and a non-linear budget set. We use Medicare Part D claims data from 2006-2009 to estimate a flexible model of consumption that accounts for non-linear budget sets, dynamic incentives due to myopia and uncertainty, and price salience. By using variation away from kink points, we are able to estimate structural models with a linear regression of consumption on coverage range prices. We then compare performance under several candidate models of expectations and coverage phase weighting. The estimates suggest small marginal price elasticities and substantial myopia; we also find evidence that salient plan characteristics impact consumption beyond their effect on out-of-pocket prices. A hyperbolic discounting model which allows for salient plan characteristics fits the data well, and outperforms both rational models and alternative behavioral models.