We use a segmentation arithmetic of self-adaption to deal with the spectral images of traditional Chinese medicine (TCM) obtained from TCM spectral detection. Through the motion detection, automatic selection of differential image and statistics of each spectral curve′s central-wavelength, the pixel points of TCM image are classified and the spectral image of TCM is divided into different areas automatically. Then spectral information is extracted from different areas of TCM′s spectral image and the effect of background noise on experimental result is eliminated obviously. As an example, images of Coptis chinensis and its mixed powder are processd with this arithmetic. The experimental results indicate that this arithmetic can automatically pick up effective areas in a precise way. It can better eliminate the interference from noises and would not produce any useless area.