Abstract: In order to reduce the current dependence on fossil and nuclear-fueled power plants to cope with the growing demand of electrical energy, the ocean energy technologies must be improved to develop more energy. There are several types of ocean energy that can be feasible to exploit: wave energy, marine-current energy, tidal barrages, ocean thermal energy and so on. But the most promising in the short term may be wave and marine-current energy. Marine-current energy can be exploited by a marine current turbine. So how to improve the efficiency of mariner current turbine is the key research subject in ocean energy development. The key to efficiency improvement is the performance improvement of hydrofoil, which is used to establish the turbine blade. In order to improve the hydrofoil's performance, a multi-point optimization method is presented in this paper. In this method, the Bezier curve was used to parameterize the hydrofoil. The Latin Hypercube experiment design method was used to select the sample points in the design space which were used for training the Radial Basis Function neural network. The hydrodynamic performance for each sample was calculated by the computational fluid dynamic method, and then the Radial Basis Function neural network would be trained by these sample points. After the neural network had been trained, the multi-point optimization method of hydrofoil was solved by combining the NSGA-II method and the Radial Basis Function neural network. The method mentioned above was applied to the optimal design of NACA63-815 hydrofoil, and the optimization problems of the hydrofoil in three typical conditions in which the attack angle is 0, 6o and 12o were mainly studied in this paper. After optimization, two optimized hydrofoils were selected in the Pareto solution to compare with initial, which were named Optimal A and Optimal B. According to the CFD simulation, the optimized hydrofoil's performance was gotten and compared with the initial hydrofoil. By comparison, it was found that the drag coefficient of the optimized hydrofoil in the three conditions are less than or equal to the initial. Moreover, the lift-drag ratios of the Optimal A hydrofoil in which the attack angles is 0, 6° and 12° have been improve by 4.6%, 4.4% and 22.8% respectively. And the lift-drag ratios of the Optimal B hydrofoils have also been improved by 6.6%, 3.8% and 16.6% respectively. In addition, according to the comparison of the optimized and initial hydrofoil's pressure coefficient at the 12°attack angle, it can be found that the optimized hydrofoil can effectively suppress the stall phenomenon. Finally, two conclusions can be drawn from the optimal results. Firstly, use the Radial Basis Function neural network to replace the CFD simulation can effectively decrease the time that the optimization cycle spent. Secondly, the hydrofoil optimization problem is a multi-solution problem. The optimized hydrofoil does not have the only one geometry. The optimized hydrofoil can have different geometry just like Optimal A and Optimal B. Moreover, both Optimal A and Optimal B hydrofoil have better performance than the initial one. The optimal result also confirms the feasibility and the theory validity of this optimization method.