Since impeller shape has great influence on hydraulic performance of a torque converter, a multi-objective optimization method based on non-dominated sorting genetic algorithm II (NSGA-II) has been used to redesign the impeller geometry. Radial basis function (RBF) is attempted to establish the surrogate models for performance responses in impeller design. A sophisticated automotive torque converter case is exemplified, which demonstrates that RBF provides a better surrogate accuracy and NSGA-II is more effective than the other methods studied. To verify the optimization results, the complete numerical characteristic curves of the torque converter with the optimized impeller are compared to the validated numerical characteristic curves of the initial torque converter. The numerical results show that the stall torque ratio and peak efficiency are increased by 3.18% and 1.4%, respectively. The results indicate a reasonable improvement in the optimal design of torque converter impeller and a higher performance using the NSGA-II method.