Clustering is one of the important technical approaches in data mining, which is widely used in knowledge discovery and machine learning. Some existing clustering methods mainly focus on either the distribution density of elements or the similarity of the morphology formed by elements in the data set, while lacks of the analysis of the intrinsic characteristics of the data. On the basis of the reasonable kernel of K-means clustering algorithm, a new clustering method is proposed in this paper and named as the K-regression clustering to realize the regression clustering. This new clustering method is based on the correlation of different components within the data set. Using this method, the data set is divided into several categories, and the data from the same subset obey a identical linear regression, while there are significant differences between any two different linear regression equations from different data subsets. Simulation results verify the rationality and validity of the K-regression clustering method.