Abstract: Color is one of the important visual features of an image．Almost all present CBIR systems use color as an essential content for matching and retrieving．Dominant colors play an important role in human vision to match images．A method to extract and represent color information of an image based on color clustering algorithm is proposed．A proper stop criterion for clustering algorithm is also presented．The features of low dimension and high precision are obtained using this method over extracting color by equal quantization in a certain color space．