The question as to whether or not any or all of the phenoxy herbicides are carcinogenic to humans continues to be evaluated. We review the evidence available from the retrospective cohort and case-control epidemiology studies. Graphs of the individual probability densities for the odds ratios from the eight case-control studies of soft-tissue sarcoma, Hodgkin's disease, or non-Hodgkin's lymphoma demonstrate gross inconsistencies which are not likely to be attributable to chance. Early studies, conducted in Sweden, had indicated strong associations, but subsequent work from New Zealand and the United States has failed to substantiate those findings. The reasons for the discordant results may relate more to methodologic problems in the earlier studies than to qualitative or quantitative differences in the exposures of the underlying populations. The retrospective cohort studies offer the advantage of having focused on occupational groups believed to have had the highest exposures, although they have been criticized as being individually too small to assess the risks of the rarer forms of cancer. Consideration of the combined cohort studies of workers exposed to the phenoxy herbicides per se provides little or no evidence of carcinogenicity. Thus, the total weight of evidence currently available does not support a conclusion that the phenoxy herbicides present a carcinogenic hazard to humans.