The paper investigates the use of image histograms as watermarks. First, the problem of exact histogram specification is addressed and a method for exact histogram specification, consistent with the human perception of brightness, is developed. Next, two watermarking techniques based on exact histogram specification are proposed. The first one directly considers image histograms as watermarks. Thus, a particular histogram is assigned as a watermark and images are further transformed to have exactly the assigned histogram. Since quite large variations in image histogram are not perceived by humans, an unlimited number of invisible watermarks can be defined for which images appear visually non-distorted. Besides, by selecting histograms which are variations of uniform histogram, the transformed images are not only uniquely marked but also enhanced. The second approach conserves, for each image, its original histogram. The watermarking procedure consists of two histogram specification transforms: a transform to the assigned watermark followed by an inverse transform to recover the original histogram. Since image recovery after a histogram specification transform is not exact, the error obtained after the two consecutive transforms is further used to track each watermark.