A steganalysis algorithm based on statistical characteristics of DCT(Discrete Cosine Transform) coefficients for JPEG(Joint Photographic Experts Group) image was proposed. The algorithm analyzes statistical characteristics of DCT coefficients, calculates a total of 8 dimensional feature vectors from each JPEG image, and using LS-SVM(Least Squares Support Vector Machines) classifier to classify cover and stego images, which can apply to blind steganalysis. The algorithm can be implemented conveniently and computed easily. The experiments results show that, the algorithm has good performance in both detection rate and computation speed, and it is also reliable to steganalysis for all kinds of JPEG images steganography.