An investigation was conducted to examine the reliability (reproducibility) of self reported occupational histories obtained from a cohort of 326 capacitor manufacturing workers who had participated in an epidemiological study relating health abnormalities to exposure to polychlorinated biphenyls (PCBs). For a subsample of the cohort (n = 164) in which occupational histories were obtained twice (in 1976 and 1979), reliability of cumulative exposure to PCBs ranged from 93.6% for the early PCB period (1947-70) to 95.7% for the late PCB period (1971-6). These respective reliabilities were lower, however, for workers who changed jobs often. Workers above the median value of a weighted job change index had early and late reliabilities of 89.9% and 83.6% respectively. Reliability is a relevant factor when calculating power or sample size during the planning stage of epidemiological studies, for interpretation or adjustment of estimates in the analysis stage, or for determination of study feasibility.