Are pictorial spaces evoked by intentionally ambiguous pictures internally coherent? If so, then we expect these to be idiosyncratic. We addressed this issue empirically. Three sheets from the second edition of Piranesi’s Carceri d’Inventione were selected for the experiment. These were compared with a conventional landscape drawing (after a Capriccio by Guardi) as a base line. The pictorial spaces were probed by way of pairwise depth-order judgments. In each image about fifty landmarks were selected, thus yielding over a thousand binary depth-order judgments per image. From such a full set of pairwise comparisons one obtains a linear depth order. This linear order again yields
postdictions for the individual pairwise judgments. Actual judgments differ from these postdictions, thereby yielding a convenient handle on the coherence of the pictorial space. Since the Piranesi Carceri are generally considered to be ‘difficult’, ‘ambiguous’ or ‘labyrinthine’, whereas a conventional landscape as the Guardi is designed for its well-defined depth structure, it is expected that they will give rise to very distinct degrees of coherence. This is indeed what we find. We studied the variation of coherence over observers and images in detail. The pictorial locations that are heavily involved in incoherent responses were identified, and the reason for their ambiguous nature traced in the structure of the depth cues provided by the artist (Piranesi).We speculate that the Carceri manage to strike just the right level of ambiguity (between total chaos and obvious structure) so as to render them visually attractive.