Causes of death of 625 subjects who died during the 4-year follow-up of a large population-based elderly cohort (Three-City study) were independently classified by the study adjudication committee and the national mortality register. The former used all available data about the cause of death (hospital records, medical data obtained from family physicians or specialists, and proxy interviews) and the latter used internationally standardized recommendations for processing death certificate data. Comparison showed a moderate overall agreement for underlying cause of death between the study adjudication committee and the national register (kappa = 0.51). Differences were found especially for cardiovascular diseases (20.6% of deaths from the study committee vs. 32.5% from the national register) and ill-defined causes of death (22.7 vs. 4%). The proportion of disagreement increased in participants dying at age >85 compared to those dying at age < or =70 (adjusted odds ratio = 2.46, 95% confidence interval = 1.10-5.49). It was also higher when the study committee used hospital record data for defining cause of death, compared to adjudication based on data obtained from proxy (adjusted odds ratio = 1.85, 95% CI = 1.09-3.14). These findings raise questions about the validity of national mortality registers in very old persons. Disease-specific causes of death, especially vascular diseases, could be overestimated in this age group.