QUESTIONS UNDER STUDY: In an attempt to analyse whether ICD-based diagnostic groups are an appropriate approach to the remuneration of Swiss psychiatric inpatient care, we investigated whether resource consumption in terms of length of stay can be predicted by diagnostic groups as well as by sociodemographic, clinical and admission-specific variables. METHODS: Data of 30,616 inpatients referred to psychiatric hospitals of a defined catchment area in Switzerland between 1997 and 2003 were analysed. RESULTS: The median length of inpatient stay is 23 days, with significant variation between and also within diagnostic groups. Patients with substance-use or adjustment disorders spent the shortest time in inpatient care, while those with an organic or eating disorder remained longest in hospital. Analyses of covariance showed that ICD-based diagnostic groups alone accounted for only 9% of the variance of the logarithmised length of stay. The amount of explained variance was significantly improved by additionally including sociodemographic, clinical and admission-specific variables. Further, investigating interaction terms alongside main effects significantly improved the explained variance to an amount of 20%. CONCLUSIONS: Diagnostic groups--even if sociodemographic, clinical and admission-specific variables are included--cannot sufficiently predict length of stay to serve as basis for the financial remuneration of Swiss psychiatric inpatient care. These results confirm findings of other international studies. Future research is needed to detect variables which adequately explain resource consumption.