Gaze following is the basis of joint visual attention. We investigated the capability of human 'receivers' to single out one of many objects, defined by the gaze of a human or computer 'sender'. Deviations from the sender's target were normally distributed and judgements were highly accurate. Accuracy of gaze following under binocular and monocular vision of the receiver did not differ, but performance was poorer when only one of the sender's eyes was visible. Two types of systematic bias could be identified: upward bias and cardinal-axis bias. In summary, human gaze following is not only very precise but also surprisingly robust to manipulations of the sender cues available for guiding the receiver's eyes.