Previous research suggests that masks disrupt expression recognition, but the neurophysiological implications of this phenomenon are poorly understood. In this study, 26 participants underwent EEG/ERP recording during the recognition of six masked/unmasked facial expressions. An emotion/word congruence paradigm was used. Face‐specific N170 was significantly larger to masked than unmasked faces. The N400 component was larger for incongruent faces, but differences were more substantial for positive emotions (especially happiness). Anterior P300 (reflecting workload) was larger to masked than unmasked faces, while posterior P300 (reflecting categorization certainty) was larger to unmasked than masked faces, and to angry faces. Face masking was more detrimental to sadness, fear, and disgust than positive emotions, such as happiness. In addition, mask covering did not impair the recognition of angry faces, as the wrinkled forehead and frowning eyebrows remained visible. Overall, facial masking polarized nonverbal communication toward the happiness/anger dimension, while minimizing emotions that stimulate an empathic response.