Abstract: This study developed a novel dimension-reduction method based on sensitivity analysis of reservoir operation periods to solve the problems of dimension disaster and local optimum trapping in calculation of complicated multi-objective reservoir optimal operation. In this method, different types of water-supply rule curves in the operation periods (as a time dimension) of one year are disturbed uniformly so as to obtain the sensitive degree to objective functions for each of the periods. Then, the positions (water level or water storage) of these rule curves in the sensitive operation periods are taken as variables to be optimized by iterative searching. After certain steps, the iteration is switched to a finer global searching. The results show that the method has advantages in search efficiency and high-quality solution and improves applicability of water-supply rule curves.