The paper analyzes the factors influencing the voltage drop of the rural middle-voltage power network. Considering the self-learning ability, associative memory and approximating nonlinear mapping of neural network, a new method to estimate voltage drop of middle voltage distribution network is proposed based on BP neural network group. As the large number of sample data of voltage estimation and the complexity of classification are easy to lead to difficult network convergence, the BP network group structure is proposed. According to the specific requirements, samples are classified and input to each BP subnet to complete the single training as well as parallel training and tests. According to impact factors of the voltage drop and the actual parameters of distribution network, the neural network input and output characteristics are identified. The samples are classified according to the load distribution type, thus the complexity of BP network training is reduced and the training efficiency is improved. The simulation results of actual examples show that the proposed method is effective and feasible for estimation of voltage drop.