This paper presents a new method for distribution service restoration. The objective of service restoration is to construct a network in which out-of-service areas are minimized. This proposed method makes use of TS (Tabu Search) that is one of meta-heuristics for solving a combinational optimization problem efficiently. TS has better performance in terms of solution quality and computational time. However, it is often inclined to be time-consuming in dealing with a large-scale problem. In particular, the creation of solution candidates requires heavy computational time as the problem size becomes larger. To allocate the problem, this paper introduces probabilistic neighborhood into TS so that computational efficiency is improved. By probabilistic neighborhood, we mean that a part of solution candidates is selected in a probabilistic way. The proposed method is successfully applied to the modified IEEE 34-node distribution system with three distribution generation units. The proposed method allows system operators to speed up computational efficiency of distribution system service restoration.