Confidence in dose-response relationships arising from occupational studies is dependent upon the reliability and validity of relevant exposure and morbidity estimates. Often self-reported occupational histories are a surrogate for direct exposure measures, and self-report of symptoms serves a similar purpose in place of more objective measures of morbidity. Unfortunately, there are few methods for validating either of these sources of information. This study attempts to evaluate the reproducibility of a survey instrument, and resulting cumulative exposure estimates, utilized in a study of work-related respiratory morbidity in the aluminum industry. The survey instrument comprised measures of pulmonary function, atopic status, smoking, occupational history, and respiratory symptoms. A Task Exposure Matrix was used to estimate exposure categories and cumulative exposure scores. Two groups of employees were administered the survey instrument on two separate occasions. Group A (n = 74) completed their two surveys more than ten weeks apart and Group B (n = 43) completed their surveys less than two weeks apart. Reproducibility was assessed using Cohen's Kappa for categorically measured outcome measures and intraclass correlation coefficients for continuously measured outcomes. Agreement across most variables, for both groups, was generally high. Repeatability of self-reported respiratory symptoms ranged from 70 to 98 percent (Group A) and 88 to 100 percent (Group B). Cumulative exposure scores were highly reproducible, despite some discrepancies in the self-reported occupational histories across interviews, with most of the calculated intraclass r scores exceeding 0.8. These results give strength to studies reporting dose-response relationships derived from similarly collected data. Reproducibility of the survey data was enhanced by use of standardized questionnaire material, consistent interview structure, and calibrated equipment for objective observation. In particular the authors recommend use of a company job dictionary to optimize reproducibility of self-reported occupational information.