Problems associated with the use of epidemiological evidence to evaluate the carcinogenicity of metals are, in the main, those associated with epidemiological enquiry in general. There are, however, some additional difficulties that arise from the variety of ways in which exposure to metals can occur, the variety of metal species and compounds, and the frequency of mixed exposure. We shall, therefore, review some of these problems, and the ways in which they can be overcome, before reviewing the evidence relating to individual elements.One of the greatest difficulties is to obtain a population for study that is large enough to distinguish the effects of a specific hazard from the effects of the random variation of small numbers or alternatively to be confident that no material risk exists, when negative results are obtained (the problem of statistical power and its relation to sample size). These cannot always be overcome by increasing the size of the population, as the number of people exposed to high concentrations of the agent may be small and extension of the study to include larger numbers will only dilute the results, if the additional subjects have not been intensively exposed. In studies of cancer, there is also the problem that the first ten years after exposure to a carcinogen seldom provide much evidence of risk. Longer periods, even 30 years or more, may be required before the existence of a risk can be properly assessed, and the inclusion of large numbers of subjects who have been observed for only a short period will serve to dilute the results still further. The analysis should, therefore, be identified whenever possible, both by intensity of exposure and by length of time since exposure began.The exact nature of past exposure at the workplace is often difficult to determine and, in the case of exposure to metals, may be quite complex. August 1981 information may be available on the previous exposure of individual workers, and comparisons may be practicable only between groups of workers employed in particular occupations or on specific processes. Sometimes, when measurements of the intensity of past exposures are not available, these may have to be estimated from the intensity of current exposure in the same occupation or the length of employment may have to be used as a substitute for measurement on the assumption that exposure has been constant over time. Exceptionally, however, if exposure was particularly heavy at a time when labor turnover was high, duration of employment can be dissociated from intensity of exposure and paradoxical results may appear to be obtained.A special problem in the study of metals is that mixed exposures to more than one metal and to different valencies or oxidation states of the metal under investigation are frequent, and precise specifications of individual components of the ambient pollution may not be available because of limitations in the analytic method. Changes in the nature of the process and in the consequent exposure patterns may also have occurred over t...