We conducted two experiments (total N = 81) to investigate the basis for the large-scale horizontal-vertical illusion (HVI), which is typically measured as 15%–20% and has previously been linked to the presence of a ground plane. In a preliminary experiment, vertical rods of similar angular extents that were either large (4.5–7.5 m) and far, or small (0.9–1.5 m) and near, were matched to horizontal extents in a virtual environment by adjustment of horizontal gaps or rods. Large/far objects showed a larger HVI (∼13%) than did small objects (∼7%), as has been shown before, but the horizontal gap normally used to measure the large-scale HVI was not the source of the larger bias. In the second experiment, we found that simply separating the comparison rod in depth from the vertical rod (thus forcing an evaluation of size at a distance) was sufficient to produce a large HVI (17%), even with small rods. The results are interpreted in light of evidence that the large-scale HVI is dependent on ground plane orientation and may be related to differential angular expansion in the visual coding of elevation and azimuth.