Aiming at the problem that the positioning precision of received signal strength based indoors positioning methods for visible light communication system is low, a new indoors positioning method for visible light communication system based on deep neural networks is proposed. In this method, the visible light channel estimation technique is adopted to measure the indoors distance, so that the problems of insufficient stability and reliability of the received signal strength are resolved. Besides, a deep neural network is designed to learn the distribution characteristics of the distance vectors of the photodiode in offline phase, in order to avoid the problem that the instable light signals lead to error growth. In online phase, the target is positioned based on multiple distance vectors, thus the positioning precision can be improved further, at the same time, the time efficiency meets the requirements. Simulation results show that, in the indoors scenario, the proposed method achieves better positioning precision than traditional triangulation methods and received signal strength based positioning methods.