Steganography is the art of secret communication [1, 2]. Its purpose is to hide the presence of information, using for example images as covers. After embedding the secret message into the cover image, a stego-image is obtained. While steganography algorithms create stego-images that are perceptually natural, we questioned if they are statistically natural [3, 4]. We show that stego-images violate recent models of natural images, and discuss the implications of this both in the art of steganography and in the mathematical modeling of natural images.