This paper presents an unsupervised texture segmentation method based on wavelet analysis and mean shift. The adaptive multiscale segmentation is realized by applying mean shift cluster algorithm to features which generated by wavelet pyramid. The unsupervised texture segmentation method does not require either training or prior knowledge of the number of textures. With a proper strategy, those features are propagated through finer scales adaptively. The center of a homogenous texture is analyzed by using features in coarse resolution, and its border is detected in finer resolution so as to locate the boundary accurately. This method has an analogy with human psychophysical measurements of image appearance. Experiments on synthetic and real images demonstrate that the proposed method leads to a successful unsupervised segmentation.