Death certificates are widely used in epidemiologic and clinical investigations and for national statistics. To examine the accuracy of death certificates for coding coronary heart disease as the underlying cause of death. Community-based inception cohort followed since 1948. Framingham, Massachusetts. 2683 deceased Framingham Heart Study participants. Sensitivity, specificity, and predictive values of the death certificate. The reference standard was cause of death adjudicated by a panel of three physicians. Among 2683 decedents, the death certificate coded coronary heart disease as the underlying cause of death for 942; the physician panel assigned coronary heart disease for 758. The death certificate had a sensitivity of 83.8% (95% CI, 81.1% to 86.4%), positive predictive value of 67.4% (CI, 64.4% to 70.4%), specificity of 84.1% (CI, 82.4% to 85.7%), and negative predictive value of 92.9% (CI, 91.7% to 94.1%) for coronary heart disease. The death certificate assigned coronary heart disease in 51.2% of 242 deaths (9.0% of total deaths) for which the physician panel could not determine a cause. Compared with the physician panel, the death certificate attributed 24.3% more deaths to coronary heart disease overall and more than twice as many deaths to coronary heart disease in decedents who were at least 85 years of age. When deaths that were assigned unknown cause by the physician panel were excluded, the death certificate still assigned more deaths to coronary heart disease (7.9% overall and 43.1% in the oldest age group). Coronary heart disease may be overrepresented as a cause of death on death certificates. National mortality statistics, which are based on death certificate data, may overestimate the frequency of coronary heart disease by 7.9% to 24.3% overall and by as much as two-fold in older persons.