The aim of this study was to assess the reliability of cancer registration data in Scotland following a major re-organisation of the Scottish Cancer Registry. From a random sample of 3500 primary cancers (excluding non-melanoma skin cancers and death certificate only (DCO) registrations) diagnosed between April and September 1997, 3175 (90.7%) had medical records available for scrutiny. Data were re-abstracted by a team of trained medical coders and compared with information registered originally. Reliability was generally high for demographic, diagnostic, and fact of treatment details, but less reliable for grade of differentiation, staging variables and dates of treatment. Some discrepancies probably arose because of differing availability of information at the time of registration compared with the time of re-abstraction. Although data quality is high overall, further improvements might be achieved through continued training, structured recording of information by clinicians in medical records, and continued exploitation of the data.