Converging evidence has demonstrated our remarkable capacities to process individual faces. However, in real-life contexts, we rarely see faces in isolation. It is largely unknown how our visual system processes a multitude of faces. The current study explored this question by using the "Fat Face" illusion: when two identical faces are vertically aligned, the bottom face appears bigger. In Experiment 1, we tested the robustness of this illusion by using faces varied by gender and race, by recruiting participants from different countries (Canadian, Chinese, and French), and by implementing different task requirements. We found that the illusion was stable and immune to variations in face gender or face race, perceptual familiarity, and task requirements. Experiment 2 further indicated that binocular vision was essential for this visual illusion. When participants performed the task with one eye covered, the previously robust illusion completely disappeared. Together, these findings revealed a visual adaptation to multiple faces in the environment: The face, which is higher in the picture plane, is perceived more distant from the viewer and appears smaller in size. And the lower face is perceived closer to the viewer and appears larger. More broadly, overestimating the size of the bottom face might act as a fundamental mechanism for social interactions, ensuring the deployment of attention to those closest to self.