A new multiagent system (MAS) design for fault location, isolation, and restoration in power distribution systems (PDSs) is presented. When there is a fault in the PDS, MAS quickly isolates the fault and restores the service to the fault-free zones. Hierarchical coordination strategy is introduced to manage the agents, which integrate the advantages of both centralized and decentralized coordination strategies. Proposed MAS is composed of zone agents (ZAs), feeder agents (FAs), and substation agents. In this framework, ZA locate and isolate the fault based on the locally available information, and assist the FA for reconfiguration and restoration. FA can solve the restoration problem using the existing algorithms for the 0-1 knapsack problem. A novel Q-learning mechanism is also introduced to support the FA in decision making for restoration. The design is illustrated by the use of case studies of fault location, isolation, and restoration on West Virginia super circuit. The results from the case studies indicate the performance of proposed MAS design.