The aims of this study were to evaluate the accuracy of AN-DRG version 2.0 codings. Ninety-two separations, covering three of the most commonly occurring AN-DRG in psychiatry, were reviewed by a psychiatrist. The AN-DRG diagnosis was then compared to that given by morbidity coders. There was agreement for 69 (79 per cent) separations and disagreement for 18 (21 per cent) separations. Recoding of the 18 separations altered average length of stay data and the funding received. The high error rate of coding stimulated changes to the training of morbidity coders and registrars.