Crop growth simulation models provided convenience for agricultural production research. However, there were many differences among the models in some of their structures and process descriptions. Thus, different models might have different simulation accuracies in different regions. An insufficient irrigation experiment of winter rape (Brassica napus L.) was conducted for six consecutive years (2009-09—2015-05) in Guanzhong Plain in Shaanxi Province. And the STICS, DSSAT and APSIM models were used to simulate the phenology dates and yields of winter rape based on the experimental data to evaluate their simulation accuracies. The results showed that the simulation accuracy of STICS was the highest among the three models with average value of absolute relative error (RARE) of 3.24%;APSIM model ranked the second with average RARE of 8.79%;DSSAT model was the worst with average RARE of 11.38%. In addition, the simulation accuracy of STICS model was the highest in both phenology dates and yields of winter rape. The simulation accuracy of DSSAT model in phenology date was higher than that of APSIM model, while in yield the simulation accuracy of APSIM was higher than that of DSSAT model. However, the simulation accuracies of the three models were all low in the season of 2012—2013 due to the low precipitation, which meant that the three models were inadequate in the simulation of drought stress conditions. In general, the accuracy of the STICS model was higher than those of DSSAT and APSIM models. The STICS model was recommended for the simulation of winter rape growth in the Guanzhong Plain.