Crop water requirement is one of the most important factors in farmland water circulatory system, its accurate estimates could help with crop irrigation management. The ability of simulating evapotranspiration (ET) for CSM-CERES-Wheat model is very important, which determines the accuracy of model output parameters. The objective was to evaluate the capability of CSM-CERES-Wheat model to simulate the cumulative evapotranspiration, daily evapotranspiration and soil water content of winter wheat (Triticum aestivum L.) in 2011—2012 and 2012—2013 growing seasons under semiarid condition. The daily ET was measured by using weighing lysimeter that installed in the experimental plot. The ability of CSM-CERES-Wheat model by using two different ET calculating approaches, i.e., Priestley-Taylor (PT) and Penman-Monteith (PM), and the simulation results were evaluated and compared. Data were obtained from two experiments with three irrigation levels that were conducted under controlled condition in a rain-out shelter in Yanling, Shaanxi Province of China during the 2011—2012 and 2012—2013 growing seasons. Results showed that the simulated cumulative ET, daily ET and soil water content of CSM-CERES-Wheat model with both PT and PM approaches had great agreement with measured values, and the simulated cumulative ET and daily ET were 5.4% and 3.4% less than those of observed values, respectively. The CSM-CERES-Wheat model simulated cumulative ET for winter wheat in two growing seasons through PT approach was less than that through PM approach, the relative difference between two approaches was ranged from -3.11% to -0.05%. The model could simulate soil water content correctly as well, in the depth of 0~20cm soil layer, the RMSEn between simulated results based on two ET approaches and observed values was 39.38%, which was relatively high and not good, but below the 20cm soil depth, the RMSEn between simulation results that based on two ET approaches and observed values was less than 23.1%, and the simulation results in depth of 40~60cm soil layer got the best agreement with the observed values. The model with both PT and PM approaches simulated the above-ground biomass and grain yield accurately as well. The RMSEn of final above-ground biomass for two growing seasons were 13.57% for PT approach and 22.76% for PM approach, and RMSEn of grain yield were 11.80% for PT approach and 15.42% for PM approach, both of the simulated results had good agreement with observed values. The model using PT approach provided smaller ET and higher soil water content than those derived from PM approach as compared with measured data. Additionally, the model using PT approach provided higher final above-ground biomass and grain yield than those derived from PM approach as compared with measured data. However, the model with both two approaches could simulate the total above-ground biomass and grain yield with RMSEn within 25% of measured data. Overall, it can be concluded that the CSM-CERES-Wheat model using the two different approaches, i.e., PT and PM, was able to accurately simulate winter wheat ET and soil water content. The results also confirmed that the model could be applied in arid and semi-arid areas, and it can be used as a tool for agricultural water management and supplying the decision support for winter wheat growth in Guanzhong Plain.