An optimal Kriging surrogate model based on a 5-fold cross-validation method and improved artificial fish swarm optimization is developed for improving the aerodynamic optimization efficiency of a high-speed train running in the open air. The developed optimal Kriging model is compared with the original Kriging model in two test sample points, and the prediction errors are all reduced to within 5%. Thus, the optimal Kriging model is selected for use in each iteration to approximate the CFD simulation model of a high-speed train in subsequent optimization. After that, the strong Pareto evolutionary algorithm II (SPEA2) is adopted to obtain a series of Pareto-optimal solutions. Based on the above work, a multi-objective aerodynamic optimization design for the head shape of a high-speed train is performed using a free-form deformation (FFD) parameterization approach. After optimization, the aerodynamic drag coefficient of the head car and the aerodynamic lift coefficient of the tail car are reduced by 5.2% and 32.6%, respectively. The results demonstrate that the optimization framework developed in this paper can effectively improve optimization efficiency.