Linear Frequency Modulation(LFM)-M code hybrid modulation pulse compression signal is designed in order to further improve the detection performance of radar. Classified comparison is used to research the application of such three classic neural networks as Back Propagation(BP) network, Elman network, and Radial-Based Function(RBF) network to pulse compression. Three network structures are designed, and their corresponding network algorithms are analyzed. Through the research on simulation and pulse compression output performance, it is indicated that a faster convergence rate and better numerical stability can be obtained in the pulse compression of LFM-M code signal by adopting RBF neural network, with the output main-to-side lobe of about 60 dB.