In Geographical Information Systems devoted to archeology, the selection of the most representative element of the Geographical Database is important to increase the value of city excavations. We developed, in the CReSTIC-SIC, a statistical method to select the most representative element of a dataset. This method is based on ranking statistics, but also on a dissimilarity measure between data to determine ranks. We apply it on the extraction of the most representative elements of excavation data. In the GIS about the Roman streets in Reims, the features of the information are modeled by convex and normalized fuzzy sets; hence we use a classical metric between convex and normalized fuzzy sets to measure the dissimilarity. This allows us to extract the most representative Roman street in Reims.