Many theories have been set-forth to explain how humans acquire spatial information. In many of these theories, landmarks have played an important role. For example, route knowledge is thought to be acquired by temporally stringing together a series of landmarks, as they are encountered. Moreover, anchor point theory states that landmarks are used to cluster together or spatially classify an area. Since many of the seminal theories of spatial navigation have stressed the importance of landmarks, we attempted to explore how they influence the transfer of navigational knowledge between real and virtual buildings. To address this issue, 24 normally sighted subjects explored and learned either the floor of a real building, or the virtual rendering of the same floor. The virtual rendering was displayed on a desktop computer and the movement was controlled by keystrokes. The virtual layout contained no visual information apart from corridor structure. Subjects were instructed to learn the layout and 5 specific target locations. After training, the subjects were tested by locating each of the targets in both the virtual and real layouts. Accuracy in target localization was high, ranging from 93% for tests in real buildings following training in real buildings to 64% for tests in real and virtual buildings following training in virtual buildings. To assess which factors are important for predicting navigation performance, we used the participantﾒs trajectory data to develop move and duration probabilities. We found the important factors for predicting target accuracy include familiarity measures (i.e., temporal and frequency probabilities across the shortest paths) and the search strategy that the participant selected (i.e., entropy). The theoretical implications of these results are discussed.