Statistical manifold is a manifold of parameterized probability density function. In order to improve steganalysis rate, we propose a using statistical manifold dimension reduction in this paper. The procedure of this algorithm is as follow: first, we use fisher information metric measures the difference among probability density function of image feature vector. Then, we embed characteristics to lower Eucli—dean space by dimension reduction. Finally, we use the support vector machine (SVM) as classifier. The experimental results show that this algorithm is effective to JSteg, F5, MBS1, MBS2 and nsF5 stegano-graphy algorithm.