We investigated the time course and processes in the recognition of facial expressions in peripheral vision (10.5°). Happy faces were categorized more accurately and faster than angry, fearful, sad, and neutral faces. Consistently, the N1 (90 to 130ms post-stimulus) and N2pc (200-300ms) ERP (event-related-potentials) components were more negative, and the SPWs (slow positive waves; 700-800ms) were smaller, for happy than for non-happy faces. Computational modeling revealed that the smiling mouth became visually salient earlier (95ms) than any other region, in temporal correspondence with the N1, thus showing an attentional capture by the smile. The N2pc presumably reflected the subsequent selective allocation of processing resources to happy faces. As a result, the reduced SPWs suggest that the decision process in expression categorization became less demanding for happy faces. We propose that facial expression recognition in peripheral vision is mainly driven by perceptual processing, without affective discrimination.