In order to improve the grinding performance of large-type semi-autogenous (SAG) mill, the trajectories of steel balls under different filling numbers were obtained through the test of the test bench of mill. Discrete element equivalent model of test bench was built and the trajectories of the steel balls were simulated. Test results and simulation results of equivalent model were compared. The comparisons showed that test results of trajectories of steel balls were consistent with simulation results. The main factors influencing the grinding performance and the influence law were studied. Mathematical model between factors and evaluation indexes was obtained through the quadratic polynomial stepwise regression analysis and analysis of variance. Optimal parameters combination was also obtained. The hierarchy structure and judgment matrix between factors and indexes were created by using the analytic hierarchy process. The prediction model of comprehensive evaluation index of grinding performance was established, by which the main design parameters of large-type SAG mill were optimized. Compared with the index values before optimization, the power per unit mass and total effective collision frequency between steel balls and aggregates were increased by 10.78% and 15.47%, respectively, and the maximum wear height of the lining was decreased by 10.81%. Meanwhile, the comprehensive grinding performance was increased by 17.25%.