Nobe et al. (1998), using a gaze tracking method, has investigated aspects of an anthropomorphic agent's hand gestures which human listeners were looking at in a real-time setting. The current study was conducted to evaluate further the human-anthropomorphic agent interaction, by increasing the number of human subjects involved and by focusing upon how the subjects' attention to gestures leads to a comprehension of them. Subjects gazed at and fixated on most gestures. Moreover, subjects imitated and understood most of the gestures in a reproduction test and a comprehension test. Sometimes subjects answered these tests correctly without any fixations, suggesting they might view the gestures in their near-foveal and peripheral vision and/or use the co-occurring speech information. These results were compared to the gaze behavior of gestures in a human-human interaction reported by Gullberg & Holmqvist (1999).