Optimization of short-term reservoir operation normally involves ramping constraints of outflows and water elevations at short time steps (e.g., hourly). Random search algorithms, such as genetic algorithms, have been widely used in optimization of reservoir operation. When applying random search algorithms to hourly reservoir operation, two important issues arise. The first one is the frequent violation of ramping constraints on the hourly reservoir outflows because of the random nature of the optimization algorithm. In other words, the optimization struggles to meet the ramping constraints when finding feasible solutions. The second issue is the zigzag fluctuation of the hourly decision variables as a result of the random search, which is unrealistic to implement in practice. In this study, the Savitzky-Golay smoothing filter (also known as least-squares filter) is incorporated periodically within the routine of the Nondominated Sorting Genetic Algorithm (NSGA-II). The goal of this study is to smooth out the decision variable functions without deteriorating the performance of the optimization algorithm. The performance of the proposed approach is quantified through three indexes using a multireservoir system with 3,360 decision variables as the test case. The results show that use of the Savitzky-Golay filter not only provides a solution to the two aforementioned issues, but also significantly improves the performance of the NSGA-II for hourly reservoir operation. The optimal decisions obtained using the proposed approach display similar hourly variability to decisions of actual reservoir operation.