Spatial LSB matching steganography makes the histogram of the embedded images smoother.Based on this, a new steganalytic method that can exploit smoothness changes is proposed. Smooth areas of a given image are chosen based on the value of pixel differences. A total of 15 features include smoothness of image histogram and local co-occurrence matrix, local smoothness and extreme value of pixel difference histograms and co-occurrence differences matrixes, etc., are extracted. The twice embedded process is used to eliminate the influence of different image contents. The Fisher linear discriminator is applied for classification. Results of experiments on the NRCS and UCID databases show that the proposed method has good performance in detecting LSB matching steganography from grayscale images.