In view of the extended debate and differing opinions on whether crystalline silica is a human carcinogen, we have reviewed a selection of epidemiological reports, to identify the areas of uncertainty and disagreement. We have chosen to examine the papers which in a recent review were considered to provide the least confounded examinations of an association between silica exposure and cancer risk. We also refer to a study of the mortality of coalminers very recently reported by ourselves and colleagues. We find that parts of the evidence are coherent but there are contradictions. On examination this resolves mostly into differences between types of studies. The three types of epidemiological study included are: (i) exposure-response studies, the most powerful for the confirmation of a relationship between a specific exposure and a health effect; (ii) descriptive studies in which incidence of disease in an exposed population is compared with that in a reference population; and (iii) studies of incidence of disease in subjects on silicosis case-registers. Descriptive studies frequently though not invariably suggest an excess lung cancer risk in silica-exposed workers compared with the general population, but exposure-response studies consistently fail to confirm that the cause is exposure to quartz. A single exposure-response study of cristobalite suggests a positive relation. Both sets of evidence have weaknesses. There are uncertainties on whether the excess risks in the descriptive studies are related to silica exposure or to lifestyle, including smoking habits. There are doubts on whether the exposure estimates in some of the exposure-response studies were sufficiently reliable to detect a small risk or weak association, though they are unlikely to have missed a strong effect. Studies of subjects on silicosis case registers consistently show an excess of lung cancer, but it is not clear to what extent these increased risks represent a direct effect of silica exposure, a secondary effect of the silicosis, preferential inclusion of subjects suffering from the effects of smoking, or bias in diagnostic accuracy. This not unnaturally leads to differences in opinion, exacerbated by variations in the strength of proof required by different experts. The main scientific uncertainties in the evidence are: 1. Whether, in the descriptive studies, the excess lung cancer rates in silica-exposed workers are explicable in terms of smoking habits, socio-economic class differences and inappropriate comparison populations. Better smoking information and more carefully chosen comparison populations are needed; 2. Whether the exposure-response studies could have missed a real relationship between silica exposure and lung cancer, if one exists. Many of the exposure-response studies were conducted with great care, but weaknesses, in the available data on which the exposure estimations were based, could have caused a real relationship of lung cancer and silica exposure to be missed. These studies were sufficiently powerful to demonstr...