In the field of the architectural heritage, 3D models are mot often used with regards to communication purposes. But under which conditions could they help in coping with the meanders of the architectural or archaeological investigation process? In parallel, the architectural documentation researchers base on is rarely organised with regards to what it is about - meaning edifices and their morphology. But is it out of reach to consider the edifice's shapes, and their 3D representation, as an interface to this documentation? Our proposition brings to the fore a methodological approach that aims at answering to those two questions. The research presented here is part of an interdisciplinary research programme' that focuses on the problems of documentation, representation and analysis of the architectural heritage. One of the main issues raised is how can 3D models help in visualising not only architectural shapes and forms but also what is known and what is ignored about them. We will present our ideas and experience on how 3D models can be used, when they are not thought as final results for communication goals but as visual interpretative interfaces. We will introduce three major aspects of our research, an analysis of the architectural morphology, an analysis of the documentation conservators base their investigations, and the solutions that we have implemented in a VRML environment. In this paper, We will particularly centre our presentation on the making of VRML scenes and on the capabilities of this standard in relation with an issue that goes beyond our field of experimentation but is vital in this field, this of dealing with shape uncertainty. Our experimental set is the medieval heart of the city of Krakow.