This study assessed whether severity of psychiatric disorder varies across day hospital and in-patient units according to local need, and whether severity of disorder predicts length of stay and therefore costs. Data were collected for a consecutive series of 2230 in-patients and 712 day patients using the Social Behaviour Scale (data completed by nurses) and diagnosis and Clinical Global Impression (CGI) (completed by doctors). Severity of illness of subjects admitted to in-patient units, but not to day hospitals, was associated with under-privileged area score (UPA). Length of in-patient stay is most accurately predicted by Clinical Global Impression and six other variables relating to diagnosis, demographic status and individual hospital. Improved resource allocation for mental health services could be achieved if severity of disorder was routinely collected.