To increase presence in virtual reality environments requires a meticulous imitation of human behavior in virtual agents. In the specific case of collision avoidance, agents' interaction will feel more natural if they are able to both display and respond to non-verbal cues. This study informs their behavior by analyzing participants' reaction to nonverbal cues. Its aim is to confirm previous work that shows head orientation to be a primary factor in collision avoidance negotiation, and to extend this to investigate the additional contribution of eye gaze direction as a cue. Fifteen participants were directed to walk towards an oncoming agent in a virtual hallway, who would exhibit various combinations of head orientation and eye gaze direction based cues. Closely prior to the potential collision the display turned black and the participant had to move in avoidance of the agent as if she were still present. Meanwhile, their own eye gaze was tracked to identify where their focus was directed and how it related to their response. Results show that the natural tendency was to avoid the agent by moving right. However, participants showed a greater compulsion to move leftward if the agent cued her own movement to the participant's right, whether through head orientation cues (consistent with previous work) or through eye gaze direction cues (extending previous work). The implications of these findings are discussed.