Evaporation duct is an ubiquitous natural phenomenon over the ocean and can be diagnosed by evaporation duct model. The model proposed by Paulus and Jeske and another model established by the American naval postgraduate school are the most widely used. They are called PJ model and NPS model, respectively. Two methods are used to investigate the global sensitivity of PJ model and NPS model in China Seas. The first method is based on meteorological and oceanographic observation data in China Seas. Considering the system random error caused by sensor measurement inaccuracies, the mean relative error and mean absolute error are used as criterion for sensitivity analysis. The second method, called Extended Fourier Amplitude Sensitivity Test (EFAST), takes into account the interaction between input parameters and is used for sensitivity analysis. The results show that NPS model is more sensitive to the random errors of sensors than PJ model. The mean relative errors of PJ model and NPS model are 11.43% and 14.81%, respectively. The results of global sensitivity parameter analysis indicate that wind speed is the key factor of PJ model, while all input parameter of NPS model have relatively large total sensitivity index. In addition, sensitivity analysis results confirm that wind speed is one of main driving factors for the formation of evaporation duct. These results are valuable for the selection of diagnosis models for evaporation duct, the evaluation of radio wave propagation in the marine atmospheric surface layer, and the prediction technique of evaporation duct based on numerical weather prediction (NWP) in China seas.