This thesis adresses the issue of the impact of the dynamic unfolding of facial actions on emotion perception. The theoretical section points to the limitations of previous reserach in studying emotion recognition from static photographs of prototypical facial displays. The importance of the dynamic features of naturally occuring expressions for emotional state attribution from facial cues is highlighted. The empirical section presents a new database of 200 short videos clips of spontaneous facial expressions that have been annotated with the FACS system. A judgment study is reported that shows that observers are able to agree on the emotional meaning of non prototypical displays. Temporal pattern analysis of facial actions, head and gazing behaviors help us specify the nonverbal signal used by observers to decipher the emotional meaning of facial expressions.