It should be noticed that efficiency estimated by RSM for runner B in pump mode is 2.2% lower than CFD prediction according to Table 5. However, the differences in efficiency between the optimized runners A and B are about 1.0% both by RSM and CFD prediction. Therefore, the applicability of the present optimization process should be inspected carefully. When an RSM approach is used in a numerical optimization procedure, the accuracy of the response surface determines the improvements among initial and optimized configurations. In this CUDC-907 study, the response surface was approximatively given by CFD predicted results. Complex geometry and flow patterns make dominance hierarchy very difficult to accurately calculate the performance of the pump–turbine, which is the main reason why there are some differences between the RSM and CFD prediction in Table 5. Besides the optimized runners A and B, several other runners have been developed in Ref. . Although there are still the differences in efficiency between estimated by RSM and prediction by CFD, the optimized runners with higher efficiencies given by RSM model also have higher efficiencies through CFD calculations. Therefore the optimized results are credible qualitatively.