[Objective] The purpose of this paper is to propose an improved dynamic programming (IDP) algorithm for the optimization operation of a trans-basin water transfer system with single water source but multiple recipient reservoirs to alleviate the so-called dimensionality problem. The improved algorithm is designed on the basis of the derived first-order derivative relationship among water availability and water transfer, release decisions with the assumption of non-decreasing storage distribution characteristic. Taking the Northern Hubei Water Diversion Project as a case study, the high computational efficiency and effectiveness of the proposed IDP are illustrated. [Methods] A multi-stage optimization operation model for a trans-basin multi-reservoir system is developed. Using Bellman’s principle of optimality within DP, the multi-stage optimization problem is then decomposed into a series of two-stage optimization problems. Based on the assumption of non-decreasing carryover storage distribution characteristic, the monotonic relationship among water availability and water transfer, release decisions is derived. Following that, the above monotonic decision characteristic is employed to design the searching principle of IDP, which can significantly narrow the search space with limited state combinations. [Results] Theoretical analysis results show that: (1) the optimal water transfer decision for each recipient reservoir is a non-decreasing monotonic function of the water availability from the donor reservoir, and these monotonic characteristics can also be specified between the optimal water release for each individual reservoir and the inter-basin system’s overall water availability; (2) for a specific recipient reservoir, the water transfer decision is a non-increasing monotonic function of the water availability from that reservoir, but a non-decreasing monotonic function of the water availability from any other reservoir in the system. Numerical results from the Northern Hubei Water Diversion Project indicate that the computing time from IDP is 1.9‰-3.7‰ of DP, within 8% loss of solution accuracy. [Conclusions] Although both theoretical analysis and numerical results show a desirable balance between the high computational efficiency and effectiveness from the proposed algorithm, future work can (1) extend the algorithm design with the general principle of the monotonic decision property for the multi-objective optimization operation of the complex multi-reservoir system and (2) investigate the effect of the non-decreasing storage distribution on the optimal solution accuracy of the designed algorithm.