This work describes a method for reliability improvement of power distribution system via feeder reconfiguration. The work presented here is developed based on a linearized network model in the form of DC power flow and linear programming model in which current carrying capacities of distribution feeders and real power constraints have been considered. The optimal open/close status of the sectionalizing and tie-switches are identified using an intelligent binary particle swarm optimization based search method. The probabilistic reliability assessment is conducted using a method based on higher probability order approximation. Several case studies are carried out on a 33 bus radial distribution system and also on 118 buses large-scale distribution system, which are extensively used as examples in solving the distribution system reconfiguration problem. Further, the effect of embedded generation on distribution system reconfiguration has been considered in one case scenario. The test results show that the amount of annual unnerved energy and customer's interruptions can be significantly reduced using the proposed method. Further, the reliability assessment method and the search method proposed in this work have both shown to be computationally efficient and very suitable for reliability-constrained feeder reconfiguration problems.