The current metal-working fluid exposures at three locations that manufacture automotive parts were assessed in conjunction with epidemiological studies of the mortality and respiratory morbidity experiences of workers at these plants. A rationale is presented for selecting and characterizing epidemiologic exposure groups in this environment. More than 475 full-shift personal aerosol samples were taken using a two-stage personal cascade impactor with median size cut-offs of 9.8 microns and 3.5 microns, plus a backup filter. For a sample of 403 workers exposed to aerosols of machining or grinding fluids, the mean total exposure was 706 micrograms/m3 (standard error (SE) = 21 micrograms/m3). Among 72 assemblers unexposed to machining fluids, the mean total exposure was 187 +/- 10 (SE) micrograms/m3. An analysis of variance model identified factors significantly associated with exposure level and permitted estimates of exposure for workers in the unsampled machine type/metal-working fluid groups. Comparison of the results obtained from personal impactor samples with predictions from an aerosol-deposition model for the human respiratory tract showed high correlation. However, the amount collected on the impactor stage underestimates extrathoracic deposition and overestimates tracheobronchial and alveolar deposition, as calculated by the deposition model. When both the impactor concentration and the deposition-model concentration were used to estimate cumulative thoracic concentrations for the worklives of a subset of auto workers, there was no significant difference in the rank order of the subjects' cumulative concentration. However, the cumulative impactor concentration values were significantly higher than the cumulative deposition-model concentration values for the subjects.