[Abstract] OBJECTIVES: We propose innovative reimbursement schemes based on Patient classification Systems that include adjustments for length of stay and exceptional costs. The schemes are designed to minimize undesirable effects of economic incentives. In addition, a statistical approach to compare these schemes as well as the underlying classification systems is proposed. Both the reimbursement schemes and the statistical approach use cost and length of stay data for two successive years. First year data are used to compute estimates of class cost means and to determine the next year's reimbursements. The reimbursements are then compared to the second year's observed costs. This method focuses on the predictive power of a classification system and basically differs from the usual retrospective analyses based on the proportion of explained variance for single year data. DATA AND RESULTS: The approach is applied to discharge data from 67'137 stays in 2002 and 60'701 stays for 2003 of Swiss hospitals. These stays are grouped according to five classification systems: AP-DRGs (All patient DRGs), APR-DRGs (All Patient Refined DRGs), IR-DRGs (International Refined DRGs), AR-DRGs (Australian Refined DRGs) and SQLape. We find that, when adjusting for length of stay and outliers, these systems do not differ substantially in their ability to predict the cost of hospitals stays. CONCLUSIONS: We conclude that a classification system with a large number of classes does not necessarily provide a better cost prediction compared to a simpler one. However, the payment of a fixed amount per diem that does not exceed the marginal cost per diem (e.g., 200 CHF) and the introduction of a simple correction for exceptional costs into the reimbursement scheme substantially reduces the average discrepancy between costs and reimbursements. [Authors]