Two methods used for retrospective evaluation of occupational exposures, a case by case assessment by expert and the application of a job exposure matrix (JEM), are compared using occupational histories collected for a case-control study on sinonasal cancer. The objective was to identify the main sources of discrepancies and to contribute to an optimal use of a JEM for population-based case-control studies. Comparisons were based on job periods, and were performed separately for two substances: formaldehyde and wood-dust. Job periods were classified according to the category of exposure assigned by the matrix, and to the probability and level of exposure assessed by the study expert. The sources of discrepancies were examined for job periods probably or definitely exposed according to the JEM and unexposed for the expert, or unexposed in the JEM and probably or definitely exposed to medium or high level for the expert. Such discrepancies were observed for 8% of the job periods for formaldehyde and 3% of the job periods for wood-dust. The agreement between the two approaches was better for wood-dust than for formaldehyde. The relative importance of different sources of discrepancies was not the same for formaldehyde and wood-dust. For formaldehyde a substantial part of the discrepancies was due to disagreements between the study expert and the matrix experts, which were mostly differences in threshold limits between 'not exposed' and 'definitely exposed at a low level'. Differences between experts' opinions did not explain the discordances observed for wood-dust. The presence of additional information in the questionnaire was an important source of discrepancy for the two substances.(ABSTRACT TRUNCATED AT 250 WORDS)