This study compares the effectiveness of Case Mix Groups (CMG*) groups and Refined Diagnosis Related Groups (RDRG) in reducing Canadian length of stay (LOS) variability. The effectiveness of the two case mix grouping methodologies was assessed with a common data base, 282,459 abstracts with ICD-9 CM diagnosis codes reported to the Hospital Medical Records Institute (HMRI) from January to March 1989. Death, signouts, transfers to or from acute care institutions and cases with an outlier LOS ("atypical" cases) were excluded from the analysis. HMRI utilization management reports to acute care hospitals use a data base defined in this way. On the basis of the variance reduction statistic (R2) from ordinary least squares regression analysis, CMG groups were found to be slightly more effective than RDRGs in reducing LOS variability. R2 statistics were 45.7 and 43.8 for CMG groups and RDRGs, respectively. Within subgroups of cases, CMG groups were found to be markedly more effective with the newborn/neonate group and to a lesser extent with non-surgical cases. The severity of illness categories within RDRGs did not, over all "typical" cases in the data base, yield more homogeneous groups of cases than CMG groups, which have half the number of categories. The value of tailoring severity measurement to Canadian medical practice and Canadian diagnosis coding is highlighted.