Many U.S. factory workers are exposed to formaldehyde. Although increased risks for leukemia have been found in medical workers and other professionals exposed to formaldehyde, studies in industrial workers, who are thought to have higher exposures, have shown inconsistent associations. We extended follow-up of a cohort of industrial workers to evaluate the association between formaldehyde exposure and lymphohematopoietic cancers. The cohort consisted of 25 619 workers (865 708 person-years) employed before January 1, 1966, at one of 10 U.S. industrial plants and followed through December 31, 1994. We analyzed formaldehyde exposure (peak exposure, average exposure intensity, cumulative exposure, and duration of exposure) and mortality from lymphohematopoietic malignancies using standardized mortality ratios and relative risks and 95% confidence intervals (CIs) based on Poisson regression. Statistical tests were two-sided. Among the cohort, there were 178 deaths from lymphohematopoietic malignancies. Relative risks for leukemia (69 deaths), particularly for myeloid leukemia (30 deaths), increased with formaldehyde exposure. Compared with workers exposed to low peak levels of formaldehyde (0.1-1.9 ppm), relative risks for myeloid leukemia were 2.43 (95% CI = 0.81 to 7.25) and 3.46 (95% CI = 1.27 to 9.43) for workers exposed to peak levels of 2.0-3.9 ppm and > or = 4.0 ppm, respectively (P(trend) =.009). Compared with workers exposed to low levels of average exposure intensity of formaldehyde (0.1-0.4 ppm), workers exposed to 0.5-0.9 ppm and > or = 1.0 ppm average intensity had relative risks of 1.15 (95% CI = 0.41 to 3.23) and 2.49 (95% CI = 1.03 to 6.03), respectively (P(trend) =.088). The relative risk for leukemia was not associated with cumulative exposure but was weakly associated with duration of exposure. Relative risks for Hodgkin's disease also increased with formaldehyde exposure. Exposure to formaldehyde may cause leukemia, particularly myeloid leukemia, in humans. However, results from other investigations are mixed, suggesting caution in drawing definitive conclusions.