BACKGROUND: Population-based mortality statistics are derived from the information recorded on death certificates. This information is used for many important purposes, such as the development of public health programs and the allocation of health care resources. Although most physicians are confronted with the task of completing death certificates, many do not receive adequate training in this skill. Resulting inaccuracies in information undermine the quality of the data derived from death certificates. METHODS: An educational intervention was designed and implemented to improve internal medicine residents' accuracy in death certificate completion. A total of 229 death certificates (146 completed before and 83 completed after the intervention) were audited for major and minor errors, and the rates of errors before and after the intervention were compared. RESULTS: Major errors were identified on 32.9% of the death certificates completed before the intervention, a rate comparable to previously reported rates for internal medicine services in teaching hospitals. Following the intervention the major error rate decreased to 15.7% (p = 0.01). The reduction in the major error rate was accounted for by significant reductions in the rate of listing of mechanism of death without a legitimate underlying cause of death (15.8% v. 4.8%) (p = 0.01) and the rate of improper sequencing of death certificate information (15.8% v. 6.0%) (p = 0.03). INTERPRETATION: Errors are common in the completion of death certificates in the inpatient teaching hospital setting. The accuracy of death certification can be improved with the implementation of a simple educational intervention.