Abstract: A fuzzy neural network is a learning machine that embeds a fuzzy system (i.e., fuzzy sets, fuzzy rules) into neural networks. It is crucial to determine the architecture of the networks according to the number of fuzzy sets. In previous works, the number is determined using fuzzy clustering. However, these methods is depended on some parameters which are set artificially. Thus, the methods are somewhat subjective. In this paper, an adaptive clustering method is proposed. It implements splitting technique and limits the number of samples in the intersection of two fuzzy sets to decrease the uncertainty. Finally, the results show that our algorithm brings out significant improvement over existing methods.