Research on naive physics and naive optics have shown that people hold surprising beliefs about everyday phenomena that are in contrast with what they see. In this article, we investigated what adults expect to be the field of view of a mirror from various viewpoints. The studies presented here confirm that humans have difficulty dealing with the role of the viewpoint in reflections and consistently prove that predictions are dominated by two patterns and a frontal bias. The majority of adults correctly predict that, from a central viewpoint, the space reflected in a mirror expands beyond the orthogonal projections to both its edges. For eccentric viewpoints, half of the participants expected a different (correct) behavior, while the other half also predicted, in this case, expansion at both edges. This means that, contrary to what happens in reality, they expected the mirror to show the space that is orthogonally in front of it and also beyond it, whatever the position of the observer (frontal bias). The error also persisted after the observation of a real reflection. However, this was not found to be true with windows. Performance improved when participants were asked to recognize the correct answer out of a series of alternatives (in this condition, only a quantitative error persisted). In both tasks (production and recognition of the correct response), people relied on imagination or memory and not on the application of the optical rule that angle of reflection equals the angle of incidence. (Contains 9 figures and 2 footnotes.)