The Cerebellar Model Arithmetic Controller (CMAC) is an intelligent controller like neural networks. Different form neural networks, CMAC can be regarded as one kind of “table-look-up” learning. Research shows that by including the fuzzy concept into the cell structure of CMAC, the accuracy can be significantly improved. Such an approach is called Fuzzy CMAC (FCMAC). In this study, it will be shown that FCMAC is very similar to the Neural-Fuzzy Networks (NFN) under certain conditions. In fact, if the locations of fuzzy rules in NFN are arranged to be the same as the locations of the cells in FCMAC, then we can say that NFN and FCMAC are equivalent provided that the simplified TSK fuzzy model is considered in NFN. This paper is to report our study about the learning performance comparison between FCMAC and NFN. It can be found that because FCMAC has more than one layer to improve the association while retrieving data, the generalization capability is better than that of NFN. From simulation, it is evident that the FCMAC model has faster error convergent speed and better noise tolerance.