This thesis investigated the processing of spatially unattended, that is, “ignored” faces. It has been a matter of considerable debate over the last decades in the field of selective attention, whether ignored information is excluded, or “filtered” from further analysis, or whether it can be partly or even entirely processed. Conflicting evidence regarding the stage of the putative filter process (early vs. late filter account) were recently integrated into a capacity account of selective attention, the “Perceptual Load Theory”. According to this theory, irrelevant information is mandatorily processed as long as relevant information does not completely exhaust all available capacity. Recent studies have suggested that faces might be an exception to this account, and may even be processed automatically. For example, there is considerable evidence for the idea that faces might capture attention to a greater extent than other stimulus classes. Similarly, recent studies demonstrated that task-irrelevant faces, but not task-irrelevant non-faces, may affect processing of task-relevant information under conditions of high perceptual load. However, new studies have reported face processing as being capacity limited, in that only one face can be processed at a time. This finding strongly argues against an “automaticity” account of face processing. The current thesis examined long-term and immediate repetition effects from – presumably – unattended faces. The use of event-related brain potentials allowed a more detailed insight into the neural processes underlying face processing in the near absence of general attention resources. One main aspect of this thesis was to show that face processing is by no means unlimited, i.e. “automatic”. Rather, the experiments presented here provide the first neural evidence for a separate attention module with limited capacity, which is well-suited to, or specific for, the processing of faces.