In this study, we investigate the effects on personality expression by the gaze movements (eyeball, eyelid, and head movements) of a dialogue robot for three-party dialogue interactions．From a three-party multimodal dialogue data, we first analyzed the distributions of (1) gaze targets, (2) gaze durations, and (3) eyeball directions during gaze aversion．We then created gaze models based on these distributions for two of the speakers in the dataset who were found to express distinct personality．We generated the gaze motions with the models of these two speakers in the humanoid robot CommU, which has big eyes and controllable eyeball and head movements．Subjective evaluation indicated that the personalities are differently perceived for the same utterances accompanied by the motions generated by these two models．