Abstract: The convergence of the learning algorithm for the CMAC (Cerebellar Model Articulation Controller) with general basis functions, presented by Chiang& Lin, is associated with the selected basis functions and the sample data during learning. Therefore, it is difficult for the algorithm to obtain high learning speed as well as learning convergence. In this paper, we propose an improved algorithm and prove that its convergence does not depend upon the choice of basis functions or sample data. The simulation results demonstrate that the improved algorithm has a better learning performance than C-L algorithm and general Albus algorithm.