This paper examines the link between diagnosis and length of psychiatric in-patient stay. Up to now the main method of predicting the use of in-patient services has been to use diagnosis-related groups (DRGs), primarily in the USA. Previous findings have revealed that DRGs generally predict less than 10% of variation in hospital stay. Psychiatric DRGs are considered to lack homogeneity and are too broad. Nevertheless, diagnosis, as an indicator of resource utilization, is now on the agenda in the UK and a study which examines the link between service use and diagnosis per se is called for. Altogether 5482 patients were allocated to 43 diagnostic categories (DCs). These DCs covered mental disorders, drug and alcohol-related problems, diseases of the nervous system and sense organs, and other related illnesses. Age was used in an attempt to refine the model further. Coefficients of variation were calculated for the DCs, and multivariate regression analysis was performed to gauge predictiveness. The results reveal that DCs contain extremely limited homogeneity and only predict 3% of variation in length of stay. When age group is included the results are only marginally improved, although the numbers contained in some DCs are low. Diagnosis, even when clearly defined, is a poor indicator of resource utilization.