Numerous proposals have been suggested to revise the current adjusted annual per capita cost (AAPCC) method for paying HMOs in the Medicare program. Several evaluations of the various alternatives conclude that including a prior utilization variable is the most promising alternative. In this paper, using statistical and other criteria, we compare the current AAPCC method with three different prior utilization models. The results suggest that all three prior utilization models can predict expenditures for either individuals or a group--drawn from a cross section and a risk selected sample of aged Medicare beneficiaries--more accurately than the AAPCC method. Of the three prior utilization measures, Payment Amount for Capitated Systems (PACS) predicts actual expenditures most accurately.