The binary behavior of activation function in receptive field ofconventional cerebellar model articulation controller (CMAC) affects thecontinuity of the network output. In addition, the original learning scheme ofCMAC may corrupt the previous learning data. A control scheme, whichparallely combines the fuzzy CMAC (FCMAC) and PID, is proposed in thepaper. The weights are updated according to the credits which are assigned to thehypercubers according to their learning histories and fuzzy membership degrees.The FCMAC is powerful in control time-varying processes due to the onlinelearning ability of the FCMAC. Experimental results of temperature control haveshown that the FCMAC with online learning ability can accurately follow thecontrol trajectory and reduce the tracking errors.