In this study, a new texture segmentation method is proposed for texture analysis applications. The technique makes use of an adaptive polyphase subband decomposition to analyze textural blocks and extract the discriminating feature sets for segmentation of the images. The subband decomposition treats the texture blocks as 2-d signals having correlated spectral properties and the filter banks used split this signal into several frequency regions which are decorrelated by the decomposition. The filter coefficients are not static but modified through the process for adaptively analyzing the signal. The feature sets constructed according to the filter coefficients and the error image are fed to the classifier functions. Since the method decomposes the signal optimally, the feature extraction capability is boosted. The method is tested on artificial images produced by using Brodatz textures and texture blocks cut from IRS Pan satellite imagery.