Recognition of faces and objects is impaired when illumination direction varies. Three experiments explore whether this impairment can be explained by display changes (Biederman & Bar, 1999), and whether cast shadows help or hinder face recognition. Observers judged whether two sequentially-presented faces, shown with or without cast shadows, were the same person. The faces were illuminated from the same or different directions, and were presented in the same or different positions on the screen. In Experiment 1, performance was illumination-dependent only on same-position trials, suggesting that observers used display changes. Experiment 2 tested whether this could be explained by peripheral viewing on different-position trials. A fixation cross cued each face's location, such that observers could move their eyes to view each face centrally. Performance was illumination-dependent regardless of whether position changed. In both experiments, shadows did not affect performance, in contrast to earlier findings (Braje, Kersten, Tarr, & Troje, 1998). In Experiment 3, all faces were presented peripherally without shadows. Changing the illumination direction did not affect performance. These results demonstrate that peripheral viewing, rather than display changes, can explain why changes in illumination direction do not affect performance when position changes. The results also suggest that face representations retain illumination information.