Medicare's use of diagnosis-related groups and the frequent acceptance of length of stay as an indicator of resource utilization has caused a surge of interest in the predictability of length of hospital stay for psychiatric inpatients. By constructing a weighted least squares regression model using data from the 1980 Hospital Discharge Survey, the authors were able to account for an increased amount of variance in length of stay for the major diagnostic categories of mental disorder and substance abuse for Medicare and Blue Cross/Blue Shield patients. The enhanced ability to predict length of stay is attributed to a carefully constructed data base and an increased number of predictor variables, particularly comorbidity. Knowledge of the presence or absence of a chemical dependency unit in the hospitals from which patients were discharged substantially increased the proportion of variance accounted for in the analysis.