Autistic Spectrum Disorder (ASD) and Williams syndrome (WS) are characterized by atypical social interaction and face processing, with opposite profiles. Evoked activity in the superior temporal sulcus (STS) occurring at 260ms is an eyes-sensitive neuronal marker. Here, we tested whether a double dissociation between these two syndromes is also observed at the cerebral level within faces processing network. Forty-two participants divided into three groups (diagnosed with WS, ASD or without neurodevelopmental disorders) performed a face perception task during a high-density EEG recording. In each trial, a particular area of the face was flashed under the participants fovea. First, using the predefined filter STS at 260ms, we found that the cortical sensitivity of STS to facial features, observed in neurotypical participants1, extended to patients with WS as opposed to patients with ASD. Again, this activity was eyes sensitive. Next, we used spatial regressor of the face cue map and computed, at each time point (without spatial a priori), to what extent the multichannel EEG signal could decode this regressor. We showed that, in contrast to patients with ASD, WS patients and young adults, produce a late (260ms) STS evoked response that is sensitive to socially relevant facial features (such as the eye and mouth). We also showed that the early (170ms) evoked response can be predicted by socially relevant facial features in patients with ASD and controls, but not in WS patients. In patients with WS, the decoding peak occurred only in the late component (at 260 ms) of the EEG, while, for patients with ASD, it occurred preferentially in the early component (at 170 ms). The present study confirms the previous results reported by Lio et al. and extend these findings by dissociating the time course of neural processes involved in face perception in WS and ASD patients. It reveals a dissociation among these two patients groups and suggest that although both syndromes are associated with social disturbances, their impairment at the neural level may have a different origin. These results provide essential evidence for understanding what, where and when neurocomputations are performed in our brain.