Speaker recognition system can identify a certain person using speech analysis. Recent advances in speech processing techniques improve the recognition rate. In this paper, an efficient speaker recognition system is proposed. Firstly, a KPCA-based feature selection approach is adopted to get the efficiently reduced dimension of feature vectors and improve clustering performance. Secondly, it has been known that the KFCM has a good superiority in clustering Non-linear and asymmetric samples and it can alleviate the negative influence of the noise and outliers. Thus KFCM clustering algorithm is applied on the selected feature samples to give out a series of clustering centers in feature space, which doubtless can represent the training set in a sense. An analysis is also provided by performing different experiments on the methods that influence the recognition rate. The experiment result shows that the proposed method can resolve the reduce the recognition error rate effectively. Keywords-Speaker Recognition, KPCA, KFCM,VQ,MCS