Abstract: Microbial germ is very important in fermentation industry, which is always selected and optimized in order to reduce costs. The appearance of the colony is related to its biochemical performance, and is helpful to manual selection in industrial production. If computer vision instead of man eyes is applied to germ classification and recognition, the efficiency will be improved obviously. In this paper, an algorithm for colony image texture segmentation is presented, taking green muscardine fengus as an example. The approach is based on the wavelet analysis theory. Firstly, the original image is decomposed into feature images with 2D multiresolution wavelet representation. Then, these feature images are segmented into intermediate region images, employing 1D wavelet transforming and multithresholding technique. Finally, as the result shows, by selecting the two parameters（the scale S of 1D wavelet decomposition, and the ambiguous threshold β ）, the image can be segmented very well. Especially useful information is provided for colony auto\|selection.