Avatars act as agents for users and have been widely introduced in visual communication applications, such as social networking services or on-line games. An avatar’s face should be designed to project the user’s unique personality. For this purpose, we must determine which part of the human face is dominantly gazed at while observers form personality impressions; this result might depend on the dimension of the impression being observed. We analyzed gaze distribution data on facial images obtained from observers who judged two different impression dimensions. Gaze fixation points obtained in each impression judgment trial were represented in 2D histograms. By comparing the 2D histograms and the ANOVA results, we confirmed that gaze attention is drawn to different facial features in impression judgments.