Abstract. When talking about emotion perception, the different representations of emotion carriers (face, body, voice and so on) remain unknown in human brain. In this study, we make the participants watch videos of faces, bodies and whole persons with different emotions enclosed. Meanwhile, we used functional Magnetic Resonance Imaging to fulfill data acquisition. We analyses the 3D imaging data by applying univariate analysis and multi-voxel pattern analysis. Eventually, we find that the signal changes of whole persons were different with a simple average of the signal changes of faces and bodies. Additionally, the extrastriate body area (EBA) had the highest accuracies for emotion classification and body part classification.