Abstract: In this paper a fuzzy CMAC neural network is proposed, which is composed of input layer, fuzzified layer, fuzzy association layer, and output layer. It has the similiar single layer link weights to CMAC and updates the consequence parameters of Takagi's fuzzy reasoning through BP algorithms. The proposed fuzzy-neural structure is described and the supervised learning algorithm is derived. The simulation results with a function approximation problem are shown that the proposed scheme is superior to CMAC in many aspects.