Abstract: Feature selection is an important component of pattern recognition.For unknown class label samples set，an unsupervised feature selection algorithm based on center distance ratio principle is proposed.The algorithm uses the mountain method to get the range of clustering number and estimate original clustering centers，then K-means clustering algorithm is adopted to confirm the optimal classification number of feature subset，and then center distance ratio principle is used to measure the classification performance of feature subset，moreover the feature correlation is analyzed，so the features with good class effect and low correlation are selected.