Emotional Facial Expressions (EFE) recognition is a major marker of emotional abilities. EFE recognition ability decreases with aging and is impaired in patients with mild cognitive impairment (MCI), Alzheimer's disease (AD) and Parkinson’s disease (PD). The present study aims to explore brain mechanisms related to such impairment through the analysis of evoked potentials (ERPs). Nineteen patients with amnestic mild cognitive impairment (aMCI), 20 AD patients, 13 PD patients with mild cognitive impairment (PD-MCI), 14 PD patients with dementia, 20 healthy elderly (HE) and 18 healthy young (HY) were presented with 5 different EFE (angry, happy, neutral, sad, fearful) from the Ekman's Pictures of Facial Affect. An electroencephalogram (EEG) was recorded and, in order to distinguish evoked components underlying the apparent emotional modulation of ERP differences, a principal component analysis (PCA) over both time and space was applied to the ERPs during an EFE recognition task. Several components were identified, based on their latency and scalp topography, corresponding to the N1, P1, N170, VPP, P2, P3 and LPP. Regardless of EFE, patients with PD-MCI showed reduced N1 compared to HE. Patients with PD showed a reduced N170 to angry faces compared to HE. HE and aMCI patients showed greater VPP than HY. Finally, in response to EFE, aMCI patients showed an enhanced P2 and a decreased P3 compared to HY. To our knowledge, this is the first time that such a comparison of patients with different kinds of dementia is made on the basis of a temporo-spatial PCA of the ERPs. Results suggests a deficit in the processing of unpleasant information in patients with PD and difficulties to process faces, particularly EFE, in patients with aMCI. Results also support a deficit in the allocation of attentional resources toward the stimuli in PD-MCI patients. No significant difference was observed between patients with AD and healthy participants, old or young.