By building a new switching function which can dynamically adapt itself to the change caused by switching of distributed generation, the multiple-population genetic algorithm for fault-section location of distribution network containing distributed generation, which can avoid the premature convergence of standard genetic algorithm, is proposed. The proposed method is applicable to the complicated distribution network, and it determines the direction of the power of the system to the user as the positive direction when locates fault in distribution networks. Multiple-population collaboratively searches the solution space to avoid falling into local optimal, and the condition of convergence is the optimal individual keeping algebra, which improves convergence efficiency. The simulation results show that the proposed algorithm has the correctness of fault location, and is effective and fault tolerance.