Abstract: A new method of optimal tree structure selection of wavelet transformation for image segmentation was presented. The standard pyramid-structure wavelet transform founded on the same recursive technique: only the low-pass outputs were used. It could not adjust the decomposition to accurate and efficient texture description. Although the wavelet packet transform provided a much more detailed analysis of the frequency content of a texture, it is often the case that areas which contain little or no frequency information are recursively decomposed. So the selection of optimal wavelet basis for texture characterization is very important. By introducing the validity measure for fuzzy clustering to the decision of wavelet decomposition structure, the presented algorithm simultaneously performs the adaptive wavelet decomposition and the texture feature classification, moreover it adaptively chooses the wavelet decomposition structure and depth. Compared with the wavelet packet decomposition, the algorithm reduces the computational burden, while obtains satisfactory segmentation results.