Event-related potentials (ERPs) were recorded to examine neural responses to face stimuli in a masking paradigm.Images of faces (neutral or fearful) and objects were presented in subthreshold, threshold, and suprathreshold conditions (exposure durations of approximately 20, 30 and 300ms, respectively), followed by a 1000-ms pattern mask. We recorded ERP responses at Oz, T5, T6, Cz and Pz. The effects of physical stimulus features were examined by inverted stimuli.The occipital N1 amplitude (approximately 160ms) was significantly smaller in response to faces than objects when presented at a subthreshold duration. In contrast, the occipitotemporal N170 amplitude was significantly greater in the threshold and suprathreshold conditions compared with the subthreshold condition for faces, but not for objects. The P1 amplitude (approximately 120ms) elicited by upright faces in the subthreshold condition was significantly larger than for inverted faces.P1 and N1 components at Oz were sensitive to subthreshold faces, which suggests the presence of fast face-specific process(es) prior to face-encoding. The N170 reflects the robustness of the face selective response in the occipitotemporal area.Even when presented for a subthreshold duration, faces were processed differently to images of objects at an early stage of visual processing.