A novel unsupervised multispectral texture segmentation algorithm is introduced. The textured image segmentation is based on a causal adaptive regression model prediction for detecting different types of texture segments which are present at the image. Texture segments are detected in four mutually perpendicular directions in the image lattice. Every monospectral component is checked separately and single monospectral results are combined together. The predictor in each direction uses identical contextual information from the pixel's neighbourhood and can be evaluated using a robust recursive algorithm. The method suggested can be successfully applied also to other unsupervised image segmentation applications, e.g. range image segmentation, edge detection, etc.