Models that simulate the effects of water stress on crop yield can be valuable tools in irrigation. We evaluated the crop growth simulation model EPICphase and the model CROPWAT on their ability to simulate maize (Zea mays L.) grain yield reduction caused by water stress under semiarid conditions. The simulation of evapotranspiration (ET), harvest index (HI), leaf area index (LAI), and final biomass was also evaluated. Data from three field experiments were used to test the models. In one sprinkler‐irrigated experiment, different water amounts (0–592 mm) were applied, producing a continuous water deficit. The other two experiments were flood‐irrigated and water stress was imposed at given development stages of maize. EPICphase simulated the ET with a root mean square error (RMSE) of 40 mm. The regression of the EPICphase simulated vs. measured values of HI and yield had intercepts that were not significantly different from 0 and slopes not different from 1. EPICphase overestimated the biomass in the more water‐stressed treatments (intercept of simulated vs. measured values = 5.25 t ha⁻¹) due to overestimation of LAI. Modifications of EPICphase relative to the effect of water stress on LAI growth and on the light extinction coefficient improved the simulations of LAI, biomass, HI, and yield. CROPWAT calculated maize grain yield with a RMSE of 14% but overestimated ET in the flood‐irrigated treatments . Better simulation of ET by EPICphase makes this model more consistent for calculating yield reduction due to water stress.