In this paper, the truth that the current reversible data hiding algorithms are detectable is confirmed by experimental analysis method. By analyzing the distributions of horizontal pixel difference of natural images and watermarked images using several models, we find that the horizontal difference histogram of natural image is significantly altered after being embedded secret message. Furthermore, the difference between the horizontal and the vertical difference histogram of natural image is much less than that of the watermarked image. Then the presence of hiding message can be detected according to the distance between the horizontal and the vertical difference histogram. Experimental results demonstrate that our approach is effective and efficient than the already published schemes.