Reference evapotranspiration (ETo) is a major component of the hydrological cycle. Accurate assessment of evapotranspiration is needed for water resources management and irrigation scheduling. The performance ability of gene-expression programming (GEP) and radical basis function neural network (RBFNN) was investigated for modeling ETo in weather station of Du’an for a 5-year period (2008—2012). The data set was comprised of daily maximum temperature, minimum temperature，sunshine duration and relative humidity, which was employed for modeling ETo by using FAO—56 Penman—Monteith equation as reference. GEP results were compared with RBFNN and Hargreaves models, and their performances were evaluated through determination coefficient (R2) and root mean square error RMSE. Based on the comparisons, GEP was found to perform better than RBFNN and Hargreaves models. The GEP model which can replace Hargreaves model and the GEP model without relative humidity were established. Statistically, GEP is an effectual modeling tool for successfully computing reference evapotranspiration.