Computer based perspective taking tasks in cognitive psychology often utilise static images and auditory instructions to assess online communication. Results are then explained in terms of theory of mind (the ability to understand that other agents have different beliefs, desires and knowledge to oneself).The current study utilises a scenario in which participants were required to select objects in a grid after listening to instructions from an on-screen director. The director was positioned behind the grid from the participants' view. As objects in some slots were concealed from the view of the director, participants needed to take the perspective of the director into account in order to respond accurately. Results showed that participants reliably made errors, attributable to not using the information from the director's perspective efficiently, rather than not being able to take the director's perspective. However, the fact that the director was represented by a static sprite meant that even for a laboratory based experiment, the level of realism was low. This could have affected the level of participant engagement with the director and the task. This study, a collaboration between computer science and psychology, advances the static sprite model by incorporating head movement into a more realistic on-screen director with the aim of a.) Improving engagement and b.) investigating whether gaze direction affects accuracy and response times of object selection. Results suggest that gaze direction can influence the speed of accurate object selection, but only slightly and in certain situations; specifically those complex enough to warrant the participant paying additional attention to gaze direction and those that highlight perspective differences between themselves and the director. This in turn suggests that engagement with a virtual agent could be improved by taking these factors into account.