Mortality from cancer among workers exposed to ethylene oxide (EtO) has been studied in 10 distinct cohorts that include about 29,800 workers and 2540 deaths. This paper presents a review and meta-analysis of these studies, primarily for leukaemia, non-Hodgkin's lymphoma, stomach cancer, pancreatic cancer, and cancer of the brain and nervous system. The magnitude and consistency of the standardised mortality ratios (SMRs) were evaluated for the individual and combined studies, as well as trends by intensity or frequency of exposure, by duration of exposure, and by latency (time since first exposure). Exposures to other workplace chemicals were examined as possible confounder variables. Three small studies by Hogstedt initially suggested an association between EtO and leukaemia, but in seven subsequent studies the SMRs for leukaemia have been much lower. For the combined studies the SMR = 1.06 (95% confidence interval (95% CI) 0.73-1.48). There was a slight suggestion of a trend by duration of exposure (p = 0.19) and a suggested increase with longer latency (p = 0.07), but there was no overall trend in risk of leukaemia by intensity or frequency of exposure; nor did a cumulative exposure analysis in the largest study indicate a quantitative association. There was also an indication that in two studies with increased risks the workers had been exposed to other potential carcinogens. For non-Hodgkin's lymphoma there was a suggestive risk overall (SMR = 1.35, 95% CI 0.93-1.90). Breakdowns by exposure intensity or frequency, exposure duration, or latency did not indicate an association, but a positive trend by cumulative exposure (p = 0.05) was seen in the largest study. There was a suggested increase in the overall SMR for stomach cancer (SMR = 1.28, 95% CI 0.98-1.65 (CI 0.73-2.26 when heterogeneity among the risk estimates was taken into account)), but analyses by intensity or duration of exposure or cumulative exposure did not support a causal association for stomach cancer. The overall SMRs and exposure-response analyses did not indicate a risk from EtO for pancreatic cancer (SMR = 0.98), brain and nervous system cancer (SMR = 0.89), or total cancer (SMR = 0.94). Although the current data do not provide consistent and convincing evidence that EtO causes leukaemia or non-Hodgkin's lymphoma, the issues are not resolved and await further studies of exposed populations.