This study attempts to propose an adaptive growing quantization approach for one-dimensional cerebellar model articulation controller (1D CMAC) network. Even though the target function is unknown in advance, the learning error can be still acquired and then is utilized to determine whether the input space needs to be repartitioned or not. Once the input space is determined to be repartitioned, some new knots are inserted for further quantization, and then the number of the states is increased. Therefore, the proposed approach not only possesses the adaptive quantization ability in the input space, but also has the growing feature in the number of the states. Beside, the linear interpolation scheme is applied to calculate the CMAC output for simultaneously improving the generalization ability and reducing the memory requirement. Simulation results show that the proposed approach not only has the adaptive quantization ability, but also can achieve a better learning accuracy and a faster convergence speed.