To assess the extent to which accuracy of death certificate completion varies with level of physician training and experience. In a classroom setting, subjects were presented with six written cases of hospital deaths adapted from materials from the National Center for Health Statistics and were asked to complete the cause-of-death section of the New York City death certificate. A total of 12 practicing general internists, 21 internal medicine residents, and 35 senior medical students. The underlying cause of death recorded by each participant was compared with the correct cause determined by a nosologist. Agreement and disagreement were classified as major or minor depending on concordance within the 17 International Classification of Diseases categories. Only one internist and five residents had received formal training in death certificate completion. The overall level of agreement between underlying cause of death reported by the three groups of participants and the correct cause was 56.9% for internists, 56.0% for residents, and 55.7% for medical students, although agreement varied with the type of case, ranging from 15% to 99%. If the misclassification observed in this pilot study were widespread, it would imply a substantial underreporting of mortality from both circulatory diseases and diabetes. These data strongly support the need to include training in death certificate completion as part of physician education.