Watermarks in papers have been in use since 1282 in Medieval Europe. Watermarks can be understood much in the sense of being an ancient form of a copyright signature. The interest of the International Association of Paper Historians (IPH) lies specifically in the categorical determination of similar ancient watermark signatures. The highly complex structure of watermarks can be regarded as a strong and discriminative property. Therefore we introduce edge-based features that are incorporated for retrieval and classification. The feature extraction method is capable of representing the global structure of the watermarks, as well as local perceptual groups and their connectivity. The advantage of the method is its invariance against changes in illumination and similarity transformations. The classification results have been obtained with leave-one out tests and a support vector machine (SVM) with an intersection kernel. The best retrieval results have been received with the histogram intersection similarity measure. For the 14 class problem we obtain a true positive rate of more than 87%, that is better than any earlier attempt.