Characterization of the health risks associated with occupational and environmental exposure to ethylene oxide (a gaseous sterilization agent and chemical building block) is made difficult by the limited dose-response information contained in the few published epidemiologic studies available, and the absence of toxicologic data for chronic exposures in species other than the rat. Federal regulatory agencies have relied heavily on conventional quantitative risk assessment techniques in setting revised occupational exposure standards for ethylene oxide. This paper indicates the variability in risk assessment results that can be obtained using the multistage dose-response model and a single animal study depending on the method used to characterize risk, the health endpoint selected, the use of confidence intervals, and the method used to equate animal and human exposure levels. Selection of the most pessimistic options available in each of these four areas is shown to result in a virtually safe dose being characterized as 0.005 ppb, whereas other reasonable assumptions yield a safe dose estimate of 1.3 ppm.