The viability and stability of the Medicare Part D prescription drug program depend on accurate risk-adjusted payments. The current approach, prescription drug hierarchical condition categories (RxHCCs), uses diagnosis and demographic information to predict future drug costs. We evaluated the performance of multiple approaches for predicting 2006 Part D drug costs and plan liability. RxHCCs explain 12 percent of the variation in actual drug costs, overpredict costs for beneficiaries with low actual costs, and underpredict costs for beneficiaries with high actual costs. Combining RxHCCs with individual-level information on prior-year drug use greatly improves performance and decreases incentives for plans to select against bad risks.