The ascertainment of information on past occupational exposure of study subjects is perhaps the main problem in case-control studies of occupational risk factors. Several methods have been proposed and used but little is known of their relative merits. The present study, undertaken in the context of a large ongoing case-control study of occupational cancer in Montreal, was designed to compare the costs of and statistical power to be derived from five plausible methods of data collection: 1) job titles abstracted from routine records, 2) job titles abstracted from routine records and processed through a job exposure matrix to derive exposure data, 3) job titles obtained by interview, 4) job titles obtained by interview and processed through a job exposure matrix to derive exposure data, and 5) job descriptions obtained by interview and processed by a team of experts to derive exposure data. Statistical power of the five methods was derived for 160 hypothetical risk factors, partly on the basis of empirical data from the data set and partly on the basis of some theoretical constructs. The design based on interview and expert evaluation was used as a reference, and the degree of misclassification of other methods was estimated in relation to this reference. For fixed sample size the interview and expert evaluation design was estimated to be much more costly than the others, but it provides much greater statistical power for detecting risks. Under the conditions of this investigation, this design was the most cost-effective. However, it is not clear to what extent this finding is generalizable.