A cohort of 1971 chemical workers licensed to handle ethylene oxide was followed up retrospectively from 1940 to 1984 and the vital status of each subject was ascertained. No quantitative information on exposure was available and therefore cohort members were considered as presumably exposed to ethylene oxide. The cohort comprised 637 subjects allowed to handle only ethylene oxide and 1334 subjects who obtained a licence valid for ethylene oxide as well as other toxic gases. Potential confounding arising from the exposure to these other chemical agents was taken into consideration. Causes of death were found from death certificates and comparisons of mortality were made with the general population of the region where cohort members were resident. Seventy six deaths were reported whereas 98.8 were expected; the difference was statistically significant. The number of malignancies for any site exceeded the expected number (standardised mortality ratio (SMR) = 130; 43 observed deaths; 95% confidence interval (95% CI) 94-175) and approached statistical significance. For all considered cancer sites the SMRs were higher than 100 but the excess was only significant (p < 0.05, two sided test) for lymphosarcoma and reticulosarcoma (International Classification of Diseases--9th revision (ICD-9) = 200; SMR = 682; four observed deaths; 95% CI 186-1745). The excess of cases for all cancers of haematopoietic tissue (ICD-9 = 200-208) also approached statistical significance (SMR = 250; six observed deaths; 95% CI 91-544). Focusing the analysis on the subcohort of the ethylene oxide only licensed workers, who are likely to have experienced a more severe exposure to this gas, it became evident that all but one of the observed cases of haematopoietic tissue cancers in the cohort were confined to this subgroup, enhancing the relevant SMR to 700 (95% CI 237-1637) and the SMR of lymphosarcoma and reticulosarcoma to 1693 (95% CI 349-4953).