Conventional methods for service restoration in distribution systems are undesirable because they may be time-consuming or the solution found is locally optimal. To remedy such a situation， a restoration method based on improved binary particle swarm optimization is proposed. To begin with， an equivalent load model is used to simplify the distribution network and an analytic hierarchy process（AHP） to calculate the weight of each index is then adopted in the process of forming the object function， which is more agreeable in practice than the conventional empirical method. Last but not least， the binary particle swarm optimization method is improved in two aspects: one is the selection of the inertia weight and learning factor， the other is the control of particle similarity. The results of a case study show that the method proposed has high calculating speed and convergence and can effectively solve the problem of service restoration.