Veterans of the US military service constitute a dynamic cohort that is suitable for epidemiologic investigations. However, the ability to ascertain vital status is paramount to the value of this population in studies having mortality as a primary endpoint. The purpose of this study was to estimate and compare the mortality ascertainment obtained from the Department of Veterans Affairs Patient Treatment File and Beneficiary Identification and Record Locator System (BIRLS) with that obtained from the National Death Index on a cohort of 17,118 male veterans hospitalized in 1970 or 1971. Based on the 4,246 deaths identified in this cohort between 1979 and 1988, the sensitivities of BIRLS, the Patient Treatment File, and the National Death Index were 94.5%, 33.0%, and 96.7%, respectively. BIRLS was comparable with the National Death Index regarding mortality ascertainment and allowed for central acquisition of 89% of the death certificates from the Department of Veterans Benefits. There was no difference detected in the completeness of the BIRLS file before and after passage of the Omnibus Act in 1981. The Patient Treatment File served as a complement to BIRLS, providing additional mortality data and identification of hospitals for recall of medical records. The National Death Index provided the best mortality ascertainment, but acquistion of death certificates from individual states was extremely time consuming and expensive. The authors recommend greater consideration by investigators of sampling the veteran population and utilizing available Veterans Affairs databases for large studies in which mortality is the primary endpoint of interest.