A common form of the Ponzo illusion involves two test probes of equal size, embedded in a planar linear perspective painting depicting a three-dimensional (3D) scene, where the probe perceived to be farther is judged to be larger than the probe perceived closer to the viewer. In this paper, the same perspective 3D scene was painted on three surfaces: (a) A 2D surface incongruent with the 3D painted scene (flat perspective). (b) A 3D surface with a geometry congruent with the 3D scene (proper perspective). (c) A 3D surface with an opposite depth arrangement to the 3D scene (reverse perspective). This last stimulus was bistable and could be perceived veridically, as it physically existed, or as a depth-inverting illusion. For all experiments, observers relied on perspective gradients to estimate the size of a test probe placed within the scene; objects placed in a ‘far’ position as defined by perspective cues were perceived to be larger regardless of their physical distance. Further, illusion strength was tied to retinal size; small retinal-size differences (Experiments 1 and 2) did not affect illusion strength, whereas larger retinal-size differences (Experiment 3) did play a minor role.