Abstract: China has a large population, and food security is a major event relating to people’s livelihood. Strengthening grain yield prediction can ensure food security. Based on historical data of grain yield in Sichuan from 2001 to 2015, we adopted grey system theory and set up GM (1, 1) prediction model on the basis of the weakening buffer operator, then used four error checking methods, including residual error, grade ratiodeviation, correlation degree and posteriori error to test the reasonability and the prediction accuracy. Then we applied GM (1, 1) prediction model to predicting the grain yield in future three years. Test results indicate that the grey system theory is appropriate when it is applied to grain yield prediction and has high prediction accuracy. According to GM (1, 1) prediction model, the grain yield of the year 2016, 2017, 2018 will increase by -2.11%, -0.39% and 1.21% respectively year on year. It can be concluded that the grain yield of Sichuan will increase in fluctuation in the near future.