The accuracy of cancer registration data in the East of Scotland (Tayside) Cancer Registry was audited by comparing 200 consecutive registrations (about 10% of the annual total) with the 'gold standard' of the Histopathology records. ICD codes were independently generated by a pathologist by examining final pathology reports and then compared to those codes given by the local cancer registrar. Discrepancies were graded by the pathologist and the epidemiologist according to severity. Major errors of coding were few. Minor and moderate differences in coding occurred because of the nature and structure of the coding system and the manner in which data are retrieved. The level of detail required by the Cancer Registry needs to be evaluated.