Sensitivity analysis of model parameters is a very important step in the process of model localization and regionalization. Local sensitivity analysis neglected the indirect effect of interactions among parameters on the model results, which resulted in the one-sidedness of choosing sensitive parameters. Sensitive parameters of winter wheat yield simulation in EPIC model were analyzed by global sensitivity analysis in Hengshui experimental area. The results showed that normal harvest index (HI), point in the growing season when leaf area begins to decline due to leaf senescence (DLAI), potential heat units (PHU) and maximum crop height (HMX), with a total sensitivity index exceeding 0.1, were the key parameters which effected the model localization. And planting date, harvesting date, and planting population, with a total sensitivity index above 0.1, were the most sensitive parameters which effected the yield estimation of winter wheat at regional scale. The study suggests that global sensitive analysis is more effective for model localization and regionalization than the local sensitivity analysis.