Summary: Four different texture classification methods (wavelet-based, co-occurrence matrices-based, 1D-histograms-based, and 1D Boolean model-based) are systematically compared and evaluated with respect to their performance in identifying textures from small and irregular samples. Two sets of 135 complex shape masks (symmetric and nonsymmetric) are created using Fourier shape descriptors, and used to clip out regions of images that must be classified. Two main series of experiments are carried out: one in which the clipped test image is from the same full image included in the database, and one where the clipped test image is from a different realization of the full image included in the database. In both occasions the best performing method was the 1D sum and difference histogram-based method.