A novel and efficient wood recognition method based on texture analysis is presented in this paper. Firstly, the sample images are divided into several regions after cutting from wood stereogram images. Then, more features are extracted by Gabor Wavelets through five scales and eight orientations. For getting the key points of these Gabor features, clustering and sifting operation are used to dislodge the dimmed features that extracted from the noise regions, such as cleavage region, resin canal region and so on. Finally, the Earth Mover’s Distance is used as the distance measure to compare these features by nearest neighbor classifier. The experiment on 24 species and 480 sample images shows that the recognition rate can even up to 97.5 percents, which gives a satisfactory classiﬁcation performance compared with the current state-of-the-art methods.