In this paper, a method with which to identify a national cohort of elderly twins is proposed. Record linkage algorithms were developed and applied to the > 30 million records contained in the Medicare beneficiary file. The matching algorithm for male/male pairs used race (black or white), last name, date of birth, and state of issuance of the Social Security number. Female/female and male/female pairs were selected with matching on race, date of birth, and the first seven digits of the Social Security number to compensate for the absence of maiden names. A stratified random sample of same-sex and opposite-sex white and black pairs (six groups) were selected and surveyed for determination of the actual prevalence of twins. On the basis of these results, the authors conclude that this method could identify an estimated 18,308 male/male, 7,544 female/female, and 204 male/female pairs of twins aged 65 years or more. This would be the largest sample of older twins ever assembled in the United States and represents a significant new resource for epidemiologic studies of the aging population.