Crop model has been becoming a powerful tool for agricultural water and nitrogen management and implementation of watersaving irrigation. This study was to explore the accuracy of CERES-Maize model for its simulations of summer maize growth, development, yield, and soil moisture under different scenarios of water stress. Field experiments were conducted under a rainout shelter for summer maize growing under water stresses at different growth stages in two consecutive growth seasons (2013 and 2014). The whole growth season of maize was divided into four stages (seeding, jointing, tasseling, and grain filling). Water stress occurred at every single stage, while irrigations were applied at the other three stages. Thus, there were four different levels of water stress period (D1~D4). Two irrigation levels of 70mm (I1) and 110 mm (I2) were applied according to the average rainfall during growth season of summer maize in 56 years. Consequently, there were a total of 8 treatments, with 3 replicates for each. The plots followed a split-plot experiment design. An extra control treatment with irrigation at all four stages was arranged nearby. The experimental data were used to calibrate and validate the CERES-Maize model with two parameter estimation tools of GLUE (Generalized likelihood uncertainty estimation) and PEST (Parameter ESTimation). Additionally, an overall evaluation was made with cross validation method for the prediction accuracy of the CERES-Maize model. Results showed that both GLUE and PEST had good stability and convergence for the estimation of genetic parameters of CERESMaize model. The parameters values separately estimated with GLUE and PEST were very close. However, PEST had higher efficiency since it consumed much less time than the GLUE. CERES-Maize model can precisely simulate the growth, development, yield, and soil moisture of summer maize under full irrigation condition, since the absolute relative error (ARE) and relative root mean squared error (RRMSE) values of model calibration and verification were only between 6% and 8%. Anthesis and maturity dates of summer maize were different when water stresses occurred at different growth stages, but CERES-Maize model failed to simulate such kind of phenology differences caused by water stresses. In cross-validation, model simulation errors became bigger when water stresses occurred at early stages, especially at jointing stage. CERES-Maize model failed to correctly simulate the influences of water stresses at early growth stages on the final grain yield of summer maize, which was probably caused by the underestimation of LAI under such conditions. Lower estimated LAI values then made the simulations of ET incorrect. In general, CERES-Maize model was proved to be limited to simulate the growth, yield, and soil moisture of summer maize when under serious water stresses at early growth stages. It is necessary to modify accordingly the CERES-Maize model if it will be used in the simulation of agro-ecological systems of summer maize in arid and semi-arid areas of China.