Accurate assessment of occupational history is critical in case-control studies of disease risk associated with employment. However, in some studies of rapidly fatal diseases, a surrogate or next of kin is interviewed rather than the study subject. In a unique re-interview study of subjects from a community-based case-control study originally interviewed 5 years earlier, we evaluated the level of agreement between (1) subjects and their surrogates on re-interview among those who had died since the first interview, and (2) subjects themselves and their responses at re-interview among those still living (to use as a comparison) regarding work history and specific occupational exposures. For employment start and stop dates, number of years worked, and number of jobs reported in the 1980 interview, exact agreement was poor for surrogate respondents and for self-respondents in the re-interview, with percentages ranging from 4% to 40%. Agreement was similar for surrogates and self-respondents for job and industry worked the longest, but percent of agreement among surrogates was significantly lower than among the self-respondents for job and industry worked last. Five (28%) of the kappa values for the industry in which the subject worked had high agreement (kappa > 0.75) for self-respondents and surrogates. None of the kappa statistics for materials handled by subjects in their jobs had high agreement, and 57% of the kappa statistics had poor agreement (kappa 0.0 to 0.39). This study suggests that occupational histories obtained by interview are poorly reproduced, and existing methods may need improvement to collect reliable occupational data. Further, the findings indicate that caution should be exercised when creating a job-exposure matrix based on such data.