The number of bed clays per inpatient episode, the length of stay (LOS), is a major indicator of hospital performance and a basic measure of patients' resource consumption. Hospital reimbursement on the basis of treated cases requires a system for accurately identifying case categories. Diagnosis Related Groups (DRGs) have been proposed for this purpose. An initial study to analyze variations in length of stay and resource consumption within DRGs is presented. Regression analysis of variation in ALOS for 7 DRGs, in terms of 8-10 independent variables not included in the classification scheme itself, was done. Results indicate that 30-65% of the large intra-DRG LOS variations are explainable by indicators of case complexity and severity despite the homogeneity claimed for the DRGs. For certain DRGs, such variations are also related to admission factors. Results indicate the need for more precise patient taxonomies than the ICDA-8-based DRGs.