In view of the problem that multiple targets are present in the main-lobe of the rotating radar, a method based on the idea of spatial spectrum estimation is proposed, which apples the RBF neural network theory to scanning radar. First a signal model of scanning radar system DOA estimation is presented and a high resolution DOA estimation algorithm is advanced based on RBF network. Then the lower speed of learning and convergence for conventional method is analyzed, thereafter a network learning algorithm based on fuzzy algorithm for learning vector quantization is proposed. In this method, the fuzzy weigh function of the mean of fuzzy-C and the online adaptive adjustment is adopted to determine the weigh between the input and the centre, which makes the network possess a better nonlinear approach performance and high-efficiency convergence. Both the theoretical analysis and the simulation results indicate that this network is fast and exact in estimation performance. The algorithm is effective and is of higher practical value.