In a distribution system, the system loss caused by line resistance can be minimized by changing its network configuration. For minimizing system loss, reconfiguration of the network is performed by opening or closing switches in the distribution system to achieve different combinations of switching status. A stochastic optimization technique of particle swarm optimization (PSO) based on the multi-agent system (MAS) is introduced and improved in this paper to solve the distribution system reconfiguration problem. To further enhance the efficiency of the algorithm, the hashing scheme is proposed to eliminate the repeated power flow analyses. The proposed method is verified in a 33-bus test system and compared with current heuristic methods to demonstrate its effectiveness.