Background/Aims: Beryllium exposure may lead to sensitization and chronic beryllium disease. Exposure-response relationships have been inconsistent, which may, in part, be because of lack of accurate and precise estimates of historical exposures. The purpose of this work was to create individual exposure estimates using work histories and a jobexposure matrix for a cohort of workers hired after 1 January 1994 at a beryllium production facility and surveyed for beryllium sensitization and chronic beryllium disease in 1999. Methods: Baseline exposure estimates (BEE) for the job-exposure matrix were generated using personal cassette samples (n = 4026) collected for all jobs during a sampling campaign in 1999. Historical exposure estimates (HEE) were generated for all job and year combinations by applying fractional annual change in area-sample exposure (1994-1998) to the 1999 BEEs. Reliability of the BEEs and HEEs was evaluated by comparing with a validation dataset of independently collected personal cassette samples between 1999 (n = 147) and 1994-1998 (n = 285), respectively. Precision, bias, and total agreement of BEEs and HEEs were assessed using correlations, Bland-Altman plots, and concordance correlation coefficients (CCC), respectively. Fractional annual change was compared with information on process changes and engineering controls instituted between 1994 and 1999, which were obtained from detailed monthly reports and interviews with facility engineers. Results: BEEs and HEEs underestimated measured exposures in their respective validation datasets by 4% and 25%, respectively. The CCC, which reflects deviation of the fitted line from the concordance line, showed much better agreement for both BEEs (CCC = 0.79, accuracy = 0.98, precision = 0.80) and HEEs (CCC = 0.65, accuracy = 0.99, precision = 0.67). Percent bias did not change with exposure levels, process areas, or year. The fractional annual change in exposure was consistent with process changes, such as implementation of enclosures and engineering controls reported in monthly reports for most process-areas. Conclusion: The method used for historical reconstruction of exposures was reasonably reliable and can be used in epidemiologic studies to evaluate quantitative exposure-response relationships.