<p>Previous studies of human-avatar interaction have shown that users are able to take the perspective of an avatar and act as if they were in the avatar’s place. As a result, the positions of objects are encoded from the avatar’s perspective and within the avatar’s spatial reference frame, and avatar-based compatibility effects can be observed. In the present study, we confronted participants simultaneously with two different avatars that offered conflicting visual perspectives. We asked participants to take the perspective of one avatar while performing a Simon task. In one half of the experiment, participants controlled the hands of the instructed avatar, while in the other half, they controlled the opposite avatar. We used stimulus-response compatibility effects to examine which perspective participants took, allowing us to examine whether instruction or control was more important in resolving frame of reference conflicts. Overall, we observed a compatibility effect based on the position of the controlled avatar, even when the participants were instructed to take the perspective of the opposite avatar. This suggests that the stimuli were encoded from the perspective of the controlled avatar, despite receiving conflicting instructions. The influence of action control overpowered the top-down influence of the instruction.</p>