The degree to which emotional aspects of stimuli are processed automatically is controversial. Here, we assessed the automatic elicitation of emotion-related brain potentials (ERPs) to positive, negative, and neutral words and facial expressions in an easy and superficial face-word discrimination task, for which the emotional valence was irrelevant. Both emotional words and facial expressions impacted ERPs already between 50 and 100 ms after stimulus onset, possibly reflecting rapid relevance detection. Following this initial processing stage only emotionality in faces but not in words was associated with an early posterior negativity (EPN). Therefore, when emotion is irrelevant in a task which requires superficial stimulus analysis, automatically enhanced sensory encoding of emotional content appears to occur only for evolutionary prepared emotional stimuli, as reflected in larger EPN amplitudes to faces, but not to symbolic word stimuli.