The underlying neural mechanisms of implicit and explicit facial emotion recognition (FER) were studied in children and adolescents with autism spectrum disorder (ASD) compared to matched typically developing controls (TDC). EEG was obtained from N = 21 ASD and N = 16 TDC. Task performance, visual (P100, N170) and cognitive (late positive potential) event-related-potentials, as well as coherence were compared across groups. TDC showed a task-dependent increase and a stronger lateralization of P100 amplitude during the explicit task and task-dependent modulation of intra-hemispheric coherence in the beta band. In contrast, the ASD group showed no task dependent modulation. Results indicate disruptions in early visual processing and top-down attentional processes as contributing factors to FER deficits in ASD.