Two experiments were carried out to demonstrate that newborns discriminate face-like configurations with highly similar inner features (Exp. 1) and to show that the same highly similar face-like configurations are treated as equivalent when compared with another face-like configuration with highly different inner features (Exp. 2). This evidence indicates that newborns are able to extract an unchanging aspect of face-like stimuli and to recognize their perceptual invariance. These results support the existence of a non-specific learning mechanism involved in face processing at birth. The functional properties of this general mechanism are discussed.