This paper deals with the seabed classification from textured sonar images and specially the potential of the combination of features extracted from co-occurrences matrices and directional filter bank (DFB) . The texture analysis based on the co-occurrences matrices is strongly dependant on the choice of parameter values (e.g. the distance and the angular direction for the estimation of the number of transitions). In most cases the choice is not trivial. To get representative features from textures with different spatial frequencies, a comprehensive set of co-occurrence matrices with corresponding displacements and orientation has to be computed. In this work, we investigate a non classical approach based on the DFB. The approach uses a decomposition of the Fourier spectrum into three spectral bands: low, medium and high frequencies. A subsequent analysis of the pattern isotropy is conducted by dividing the medium spectral band into small, overlapped, angular sectors. The features extracted from this process are assessed so as to determine their potential on the classification performances. First, a comparison with classification performances result given by texture features derived from grey level co-occurrences matrices (GLCM) is made. Finally the global performance of the segmentation is assessed using the spectral features, the features extracted from GLCM and the grazing angle. The Klein 5000 experimental data used in this study have been acquired by DGA/GESMA during BP 02 experiment conducted by NURC.