Effects of the orientation of a task-irrelevant avatar have been considered evidence of automatic perspective taking. Specifically, previous studies found that giving left- vs. right-sided key press responses to a target stimulus presented on the sagittal midline of the screen, accompanied by an avatar rotated by 90° from the participants’ orientation, yields faster responding on the side that matches the laterality of the target viewed from the avatar’s perspective (i.e., an avatar-based Simon effect). In previous studies, the orientation of the avatar was held constant during blocks of trials, preventing inferences regarding flexible, trial-to-trial, perspective taking. Presenting the avatar with clockwise or counterclockwise tilt, chosen randomly on each trial, we examined whether an avatar-based Simon effect would also occur under conditions of variable and uncertain avatar orientation. To further extend investigation of automatic perspective taking we intermixed trials of a mental rotation task requiring judgment whether a stimulus character was presented in normal or mirrored format. This character occurred in a central position viewed from both the participants’ and the avatar’s perspective and was rotated by 90° in the picture plane (i.e., presented in upright or inverted format viewed from the avatar’s perspective). An avatar-based Simon effect occurred. Performance in the mental rotation task was not better, however, when the character was presented in upright format viewed from the avatar’s perspective. Our results thus suggest flexible left-right coding of the position of stimulus objects according to the avatar’s orientation but cast doubts regarding more extensive forms of automatic perspective taking.