During an exposure monitoring study, 78 saw maintenance tradesmen were randomly assigned to be interviewed about their exposures using one of two questionnaire formats: open-ended and partly prompted questions about five categories of materials; and detailed prompting about 75 agents. The more open-ended questionnaire elicited fewer exposure responses overall, but more responses about agents not included on the detailed questionnaire. Composite materials and trade name products were more frequently cited as exposures than individual metals or compounds. Validity of responses was ascertained using air measurements (individual metals) or observations of the employees (composite materials). Sums of sensitivities and specificities were very low (near 1.0) for most of the metals for both types of questionnaire. For composite materials, validity improved substantially. Sensitivities with the partly promoted format (0.44-0.85) were always lower than with detailed prompting (0.80-1.00). Specificities were usually, but not always, higher with partial prompting (0.66-0.92) than with detailed prompting (0.18-0.86). Selection of questionnaire format for an epidemiologic study would depend on the likely prevalence of exposure in controls and the effects of trade-offs in sensitivity and specificity.