In this paper, a new method is proposed for distribution network service restoration. It is hard to determine the network service restoration due to the NP-hard problem. The proposed method is based on Tabu Search (TS) of meta-heuristics to evaluate better solutions efficiently. However, it has a drawback to require a lot of computational time in creating the solution candidates in realistic problems. To overcome the drawback, this paper introduces probabilistic sampling into TS to reduce computational effort in creating solution candidates. This paper applies TS with probabilistic sampling to the network service restoration and investigates the performance by changing the probability of sampling in the neighborhood. The proposed method is successfully applied to a sample distribution system.