Purpose: In order to better understand the interactions of spatial memory and visual information in spatial navigation, we have developed an ideal navigation model called ANTIE (Acme Navigation Through Indoor Environments) that uses visual and geometric information optimally during spatial navigation. By comparing humans with ANTIE, we have investigated the interactions of vision and memory during spatial navigation by testing how a building's layout size (complexity) affects human navigation performance. Method: We trained and tested subjects on eight different layouts (building floor plans) using first-person desktop virtual reality. We manipulated layout complexity by using random layouts based on a Cartesian grid that had 10, 20, 40, and 80 corridors. Training consisted of a series of three-minute explorations, each followed by a learning test in which subjects drew the grid-like layout. Subjects repeated these three-minute explorations until they drew the layout correctly twice in a row. During testing subjects were placed at a random location within the virtual environment and they were told to go to a target location within the environment in the minimum number of "moves" (where a "move" was defined as a rotation or a unit translation). Human performance was measured by a wayfinding ratio (# of ideal moves (ANTIE) / # human moves to solve the same task), a measure analogous to statistical efficiency. Results and Conclusions: Average wayfinding ratio decreased as the layout size increased: 80.1% for a layout complexity of 10 corridor segments, 72.3% for 20 corridors, 63% for 40 corridors, and 51.3% for 80 corridors. Using ANTIE, we tested whether the decrease in efficiency could be explained by an inefficient use of visual information. The analysis suggests that the reduction in efficiency is not due to an inefficient use of visual information (consistent with Stankiewicz et al., ARVO 2000) and suggests that it may be due to a limitation in spatial memory.