BACKGROUND: The aims of the present study were to validate the Physiological and Operative Severity Score for the enUmeration of Mortality adjusted for oesophagogastric surgery (O-POSSUM). METHODS: Data on patients who underwent potentially curative oesophagectomy in a tertiary referral centre for adenocarcinoma or squamous cell carcinoma of the oesophagus were analysed. The in-hospital mortality predicted by O-POSSUM was compared with the actual value by linear analysis. RESULTS: Twenty-four (3.6 per cent) of 663 patients died in hospital. The observed : predicted ratio for in-hospital mortality was 0.29. The model had a poor fit (P < 0.001). The area under the receiver-operator characteristic curve was 0.60 (95 per cent confidence interval 0.47 to 0.72); P = 0.113). O-POSSUM score was not related to the severity of complications. CONCLUSION: O-POSSUM overpredicted in-hospital mortality threefold and could not identify patients at higher risk of death. O-POSSUM needs substantial modification before it can be used for comparison of treatment outcomes between centres. Copyright (c) 2007 British Journal of Surgery Society Ltd.