Abstract: A new way to recognize the microimage of powdered Chinese medical herbs is put forward on multiple-feature basis according to their textural characteristics. It includes three analytic methods, i.e., fractal analysis, wavelet analysis and gray level-gradient joint occurrence matrix. As the foundation on which the images are efficiently recognized, the characteristic vector is formed by combining together the characteristic quantities that are picked out according to different methods as above and how important the rates they play in the classification of images. Using the k-nearest neighbour classification, the experimental results show that the way proposed is feasible and effective and provides a theoretical reference for the quantitative recognition of Chinese medical herbs.