The aim of this study was to evaluate a case-mix system to classify inpatients with mental disorders in Germany by means of self-report and expert-rated instruments. The use of case-mix systems enhances the transparency of performance and cost structure and can thus improve the quality of mental health care. We analysed a consecutive sample of 1677 inpatients with mental disorders from 11 hospitals using regression tree analysis. The model assigns patients to 17 groups, accounting for 17% of the variance for duration of stay. Patients with eating disorders had a longer duration of stay than patients with anxiety disorder, duration of mental illness of less than 3–5 years, lower levels of interpersonal problems and higher occupational position. The results showed that besides diagnosis, variables such as duration of illness and interpersonal problems are important for classifying inpatients with mental disorders. The results of the study should be critically reviewed regarding the empirical results of other studies and the appropriateness of case group concepts for inpatients with mental disorders.