Abstract: Aiming at the single consideration of the performance characteristics while the lack of the load characteristics in most studies about the reactor coolant pump (the main pump), a new impeller meridian design approach along with the radial basis function neural network and the multi-optimization algorithm was herein adopted, as a result, a new optimization strategy whose targets were the higher performance and the lower axial load was established. To verify the effectiveness of the proposed optimization strategy, it was then applied into the practical design process whose design target was the previous scaled main pump. It could be found as follows: with only three control variables and fifteen samples, the new strategy succeeded in optimizing the pump; in relative to the target pump, the efficiency of the optimal structure is increased by 0.9%, the head is increased by 0.6 m, while the axial load is decreased by about 200 N; the predicting results from the approximate model could quantitatively prove that the increasing of the head can easily lead to the increasing of the axial load.