Adopting a procedure developed with human speakers, we examined infants' ability to follow a nonhuman agent's gaze direction and subsequently to use its gaze to learn new words. When a programmable robot acted as the speaker (Experiment 1), infants followed its gaze toward the word referent whether or not it coincided with their own focus of attention, but failed to learn a new word. When the speaker was human, infants correctly mapped the words (Experiment 2). Furthermore, when the robot interacted contingently, this did not facilitate infants' word mapping (Experiment 3). These findings suggest that gaze following upon hearing a novel word is not sufficient to learn the referent of the word when the speaker is nonhuman. (Contains 2 figures and 1 table.)