A novel algorithm of analyzing the textural features of image is put forward for color difference classify of ceramic tiles. First the image of tile is pretreated. Then two level wavelet decompositions are used on the image, and the energy feature of each sub-graph is extracted. The energy contains color and textural features of image, which be used to classify tiles into different color. Tile under test is classified according to the similarity measurement, which is the similar distance between feature value of tiles under test and the one of standard tiles. Euclid distance is used as similarity algorithm. Experimental results show that this method can reflect spatial relationship of color more effectively, and improves the accuracy of the classification in the production line.