This paper proposes a two-stage distributed method for the restoration problem using multi-agent system. The first stage gets a quick solution and implements it to restore as much loads as possible based on their priorities. The second stage will be applied to faults which expected to take longer time in order to maximize the restored loads (if there are loads still not energized from the first stage), minimize the power losses, and relieve the overloaded lines for more efficient and economical operation. Distribution system is represented as a multi-agent system with hierarchical architecture. Four types of agents are included, which are Bus Agents (BAGs) for monitoring and implementing the switching actions, DG Agents (DGAs) to represent DGs as supporting sources of power, Zone Agents (ZAGs) for negotiation and decision in the first-stage of restoration, and Global Agent (GAG) for decision in the second-stage of restoration. The proposed method is implemented in JADE and simulation results of 16-Bus system are included.