Behavioral studies have shown that two different morphed faces perceived as reflecting the same emotional expression are harder to discriminate than two faces considered as two different ones. This advantage of between-categorical differences compared with within-categorical ones is classically referred as the categorical perception effect. The temporal course of this effect on fear and happiness facial expressions has been explored through event-related potentials (ERPs). Three kinds of pairs were presented in a delayed same–different matching task: (1) two different morphed faces perceived as the same emotional expression (within-categorical differences), (2) two other ones reflecting two different emotions (between-categorical differences), and (3) two identical morphed faces (same faces for methodological purpose). Following the second face onset in the pair, the amplitude of the bilateral occipito-temporal negativities (N170) and of the vertex positive potential (P150 or VPP) was reduced for within and same pairs relative to between pairs. This suggests a repetition priming effect. We also observed a modulation of the P3b wave, as the amplitude of the responses for the between pairs was higher than for the within and same pairs. These results indicate that the categorical perception of human facial emotional expressions has a perceptual origin in the bilateral occipito-temporal regions, while typical prior studies found emotion-modulated ERP components considerably later.