A summary and commentary was presented for a workshop on retrospective epidemiological occupational exposure assessment. While it was difficult for an epidemiological study to demonstrate a dose response relationship, a negative result did not necessarily demonstrate lack of causation. Inherent biases resulting in no response when one actually existed were discussed. These included the healthy worker effect, latency, lack or poor quality of past exposure data, routes of exposure, synergism, job classification, amount of exposure, estimate error bounds, work practice variability and failure to follow up temporary or contract workers. In discussing data interpretation, it was noted that about 30 times more data was required to prove safety than to prove a hazard. Interpretation of positive results was discussed with reference to setting allowable exposure limits based on imprecise exposure estimates and political and economic ramifications. Problems of epidemiological research associated with the workers' right to know and preventive measures were discussed in terms of time required to obtain accurate study results. The author concludes that improved exposure estimates will yield improved study conclusions and that there is a need for a bolder attitude in the area of preventive epidemiological research in order for efforts to have real impact.