A statistical methodology based on the Cox proportional hazards model (a survival time analysis method), an alternative to the approach underlying DRGs, is presented. The method is used to obtain an estimate of the length-of-stay (LOS) distribution of a patient incorporating either patient-specific or hospital variables. A percentile of the distribution chosen to minimize prediction error serves as the assigned LOS. Absolute deviation is used as the loss function both to determine the choice of a predicted LOS and to examine how well the scheme works. Multiple assignment schemes may also be developed from this approach. The results of the method, tested on a national probability sample of 4,608 psychiatric patients treated in psychiatric units of general hospitals, suggest that with respect to average absolute deviation, the proposed methodology may provide a scheme that is superior to the present DRG scheme. For the sample, the average percent improvement obtained using the median of the estimated LOS distribution as the predicted LOS over the sample mean of the DRG group is 19%. A two assignment strategy results in average improvements over DRGs of 43%.