It is generally accepted that diagnosis-related groups (DRGs) for alcohol, drug, and mental disorders are inappropriate for inpatient prospective payment. To address this issue, the Veterans Administration (VA) supported a project to construct alternative classes that are more clinically meaningful, more homogeneous in their resource use, and that account for more variation in resource use among psychiatric and substance use cases than existing DRGs. This paper reports on this project. Using a data set containing universally available discharge data plus behavioral, social, and functional information obtained by a survey of 116,191 discharges from VA psychiatric beds, and with AUTOGRP as the classifying algorithm, a classification system was formed. Twelve psychiatric diagnostic groupings (PDGs) were identified, analogous to major diagnostic groups in the DRG system. Within each PDG, from 4 to 9 terminal groups of Psychiatric Patient Classes (PPCs) were formed and validated. The 12 substance abuse PPCs explain greater than 31% of the variation in length of stay; for the mental disorder PPCs the variance explanation is greater than 11%, a substantial improvement over DRGs that, for the same data set, explain less than 2 and 3%, respectively. With the addition of only 5 variables beyond those presently included in discharge data sets, greater precision for payment purposes can be achieved. Implications for adoption of this classification system are discussed.