Data on exposure to formaldehyde from a cohort study of 26,561 workers first employed between 1934 and 1965 in 10 US plants producing or using formaldehyde were used to compare the classification of workers by different estimates of exposure, including duration of employment, duration of exposure, level for job with highest 8-hour time-weighted average exposure, estimated highest peak exposure, cumulative exposure, and average exposure. Measures of duration (employment and exposure) were highly correlated, as were average exposure and highest 8-hour time-weighted average exposure. Moderate correlations were seen between cumulative exposure and either average, highest 8-hour time-weighted average exposure, peak, or duration of exposure, between average and peak exposure, and between highest job and peak exposure. Average exposure, however, showed little correlation with duration of employment, duration of exposure, or peak exposures. The degree of similarity of these estimates of exposure also varied by plant. These results indicate that various measures of exposure rank subjects differently along an exposure gradient and that selection of an inappropriate measure could mute exposure-response gradients. Investigators should use several exposure measures in analyses whenever the mechanism of action is poorly understood in order to avoid false-negative findings.