For a cohort study of chlorophenate-exposed sawmill workers, historical exposures from the 1940s to the 1980s were estimated by teams of 9-20 employees (each interviewed individually). The mill histories were divided into eras when jobs and exposures were relatively stable. Raters with at least 5 years of work experience in an era were asked to estimate the frequency and duration of exposure for each job in the mill. Reliability measures for these estimates were calculated for each of the 39 mill and time period combinations, using the individual intraclass correlation coefficient (ICCind) to assess agreement between raters and the group intraclass correlation coefficient (ICCgroup) to assess the stability of the mean estimates of exposure. ICCind were low, with means that increased from 0.24 to 0.35 over the five decades. ICCgroup were considerably higher (means increasing from 0.78 to 0.88 over time), indicating that the number of raters used in this study was sufficient to produce stable average estimates of exposure throughout the study period. These data confirm the intuitive expectation that reliability of exposure estimates decreases when reconstruction of conditions in the distant past is required, and therefore that the random component of exposure misclassification is a greater threat to validity in these earlier time periods.