In order to deal with large-scale of candidate measures for reliability improvement of medium-voltage distribution network, a cost-effectiveness based two-stage optimal selection methodology for reliability cross-domain improvement strategy is proposed. Firstly, based on multi-source data such as asset information management system and reliability management system of power grid company, the refined reliability evaluation of distribution network is carried out. Then, the feeder is used as the implementing unit to list all the reliability improvement measures, which are divided into three dimensions of outage frequency, outage time, and outage impact user range. The empirical logic function is used to screen the improvement measures. Finally, the boundary conditions of the refined modeling measures are determined, and the reliability improvement measures are further sorted and screened based on the reliability marginal benefit. The cost-benefit optimization under the designated capital investment or reliability target is achieved. The practicability and effectiveness of the proposed method are verified by an example of medium voltage distribution network in a county-level power supply bureau in western China.