The Savannah River Site (SRS) is one of the largest facilities in the nation's nuclear weapons complex. To date, little information has been published regarding radiation risk estimates derived from epidemiological studies of SRS workers. As part of an ongoing epidemiological cohort study of SRS workers, we have assessed the suitability of the Site's personnel radiation dosimetry information for use in epidemiological analyses. This paper provides information on historical dosimetry methods, recording practices, and the completeness of computerized dosimetry information for workers employed at SRS during the period 1951-1989, when the site was operated by the du Pont Company. The study includes 18,883 workers hired at SRS between 1951 and 1987 who were employed for at least 90 days. Documents relating to external radiation dosimetry methods were reviewed, recorded doses were examined to evaluate recording practices, and the completeness of monitoring was assessed by comparing employment history and computerized dosimetry records, and by implementing a "nearby" procedure for estimating values for missing annual dosimetry records. Dosimeter technology evolved over this period from two-element film dosimeters to multielement thermoluminescent dosimeters. Dosimetry measurements were recorded consistently in 0.05 millisievert (mSv) increments. Prior to 1973, recording thresholds of 0.10-0.15 mSv were used while from 1973 to 1989 a recording threshold of 0.05 mSv was used. We abstracted nearly 3 person-Sv of dosimetry information that was available in hardcopy but not in computerized format. The collective dose from the computerized and abstracted records totaled 512.1 person-Sv. A "nearby" method was used to estimate dose values for 13,812 employment-years for which dosimetry information was not available. The average estimated value was 0.6 mSv and the assigned collective dose derived via the "nearby" procedure was 8.7 person-Sv. The consistency of dosimetry practices at SRS and the completeness of historical dosimetry records are supportive of their use in epidemiologic research.