Correct retrospective assignment of subjects to an exposure category is affected by a variety of problems: 1) lack of an objective lifetime measurement; 2) dependence upon the accuracy and thoroughness of the job description; 3) heavy reliance upon the knowledge of experts. The aim of the study was the quantification of the performance of a job exposure matrix (JEM) in evaluating solvent exposure, using expert judgements as the reference method. The sources of discrepancies between the two methods were analysed within the framework of two community-based case-control surveys. One included 765 cases of bladder cancer (BC) and 765 controls, the other 298 cases of glomerulonephritis (GN) and 298 controls. The JEM had been set up previously for a case-control study on laryngeal cancer and is based on 4000 discrete job titles. Comparison between the JEM and expert exposure evaluation was carried out for 2736 job periods in the BC study and 929 in the GN study. Categories of exposure for both experts and JEM were dichotomized, using different cutoff points for exposure and non-exposure. Prevalence of exposure as assessed by the experts was twice as high in the GN study (19%) as in the BC study (10%), showing the importance of the questionnaire design and of the inclusiveness of the definition of exposure. Sensitivity of the JEM vis-a-vis the experts was low (23-63%), whereas specificity was rather high (87-98%). The best concordance between the two methods was obtained with a specific dichotomy from the JEM and a narrow definition of exposure by the experts. Bias and loss of power resulting from JEM misclassifications were calculated with a theoretical population odds ratio of 3 and an exposure prevalence of 10%. If the experts' classification of the subjects according to exposure is assumed to be 100% correct, using the JEM led to a bias in estimating the odds ratio, ranging from 1.5 to 2.1, and to a loss of power equivalent to a reduction in the number of subjects by a factor of 5 to 10. Analysis of systematic discrepancies between exposure assessments of the experts and the JEM showed that they were clustered with some job categories and arose from different sources: 1) inadequate job descriptions, related to the codification system adopted and necessitating the gathering of information at the individual level; 2) true disagreements between JEM and experts regarding the definition of solvent exposure.(ABSTRACT TRUNCATED AT 400 WORDS)