OBJECTIVE: To estimate the proportion of seniors with dementia from three independent data sources and their agreement. DATA SOURCES: The longitudinal Asset and Health Dynamics among the Oldest Old (AHEAD) study (n=7,974), Medicare claims, and death certificate data. STUDY DESIGN: Estimates of the proportion of individuals with dementia from: (1) self- or proxy-reported cognitive status measures from surveys, (2) Medicare claims, and (3) death certificates. Agreement using Cohen's kappa; multivariate logistic regression. PRINCIPAL FINDINGS: The proportion varied substantially among the data sources. Agreement was poor (kappa: 0.14-0.46 depending upon comparison assessed); the individuals identified had relatively modest overlap. CONCLUSIONS: Estimates of dementia occurrence based on cognitive status measures from three independent data sources were not interchangeable. Further validation of these sources is needed. Caution should be used if policy is based on only one data source.