Abstract: The concept of near-ness in the shared nearest neighbors(SNN) is given an embodiment and an algorithm for determining initial cluster centers in the K- means iteration (KMT) is designed. The modified SNN (MSNN) and the modified KMI (MKMI) are merged to from an unsupervised combined clustering algorithm (CCA). For comparison, the KMI, MKMI, and CCA were applied to classify two groups of speech data. The results show that the performances of the MKMI and CCA are respectively 57% and 108% higher than that of the KMI, and MSNN is a good pre-classifier.