Detection of hidden messages in images is of great importance for both the network information security and the improvement of security of steganographic algorithms. Based on the statistical observations on the difference histogram of images, a new steganalytic technique capable of a reliable detection of the spatial LSB (least significant bit) steganography is proposed. Translation coefficients between the difference histograms are defined as a measure of the weak correlation between the LSB plane and the remained bit planes, and then used to construct a classifier to discriminate the stego-image from the carrier-image. The algorithm can not only detect the existence of hidden messages embedded using LSB replacement in images reliably, but also estimate the amount of hidden messages exactly. It has a distinct physical meaning and can be implemented conveniently. Experimental results show that for raw losslessly stored images, the new algorithm has a better performance than the RS (regular singular) steganalysis method and improves the computation speed significantly. The new approach is also applicable for color images.