A method of texture analysis based on the using Kullback-Leibler divergence is discussed. A digital image is described by a discrete probability distribution. We consider a group of direct multifractal transforms relating to the distribution and for two given images calculate a vector of Kullback-Leibler divergences between the initial distributions and their direct multifractal transforms. The method is demonstrated on the example of the Serpinsky carpet. It proves to be appropriate for textures from different classes but having similar structures and may be applied for classification of biomedical preparation images.