This paper proposes a multi-agent approach to power system restoration. The proposed system consists of a number of bus agents (BAGS) and a single facilitator agent (FAG). BAG is developed to decide a suboptimal target configuration after a fault occurrence by interacting with other BAGS based on only locally available information, while FAG is to act as a manager in the decision process. The interaction of several simple agents leads to a dynamic system, allowing efficient approximation of a solution. Simulation results have demonstrated that this method is able to reach suboptimal target configurations, which are favorably compared with those obtained by a mathematical programming approach.