A model for estimating exposure to ethylene-oxide (75218) in a retrospective cohort mortality study was developed. The model based on a linear regression procedure was intended to estimate ethylene- oxide exposures by facility and year for a NIOSH cohort mortality study of workers in the medical supplies and spice industries. Ethylene-oxide exposure data were collected during walkthrough surveys of 36 companies and in/depth surveys of two companies. A total of 230 data points were used to develop the model. Twenty three independent variables were examined for their potential association with the logarithm of the average exposure expressed as parts per million (ppm) ethylene-oxide. Seven variables, year, volume of sterilizer or treatment vessel in cubic feet, product age, exposure category, product type, facilitated product aeration, and rear exhaust valve in the sterilizer or treatment vessel, and two interactive terms, exposure category times product type and exposure category category times year, explained 90 percent of the variance in the data. When used to estimate exposure concentrations for 46 other data points not used in developing the model, the average bias was -0.5ppm and the standard deviation was 5.6ppm. An evaluation of the same test data by a panel of experienced industrial hygienists yielded a bias of -2.9ppm and a standard deviation of 5.2ppm. The authors conclude that the model explains 90 percent of the variance in the ethylene-oxide data while using only a relatively few terms.