We aimed to develop a method to classify those errors in the completion of death certificates arising from misunderstanding of the certification process. We reviewed 430 Western Australian death certificates registered in March 1990 using a method devised to differentiate between major and minor errors. Major errors were found in 16 per cent of certificates (95 per cent confidence interval 12 to 20 per cent). The error rate did not vary significantly between city and country areas, nor between teaching hospitals and other locations. The method was validated on a separate sample of 120 certificates. It has potential as a tool to monitor a critical but neglected step in the production of mortality statistics and to assess understanding of the process of death certification.