Facial expressions are an important part of human communication. However, children with autism spectrum disorders (ASD) are often suffering from difficulties of understanding non-verbal cues and form appropriate responses. Traditional approaches including labeling formatted photographs of human facial expressions from a third person's perspective could help them learn and improve such skills. Yet such training systems are often in lack of real time feedback. As Optical See-Through (OST) Augmented Reality (AR) headsets possess the advantages of near-eyes display and better depth alignment between virtual renderings and the environment, we decided on an OST AR approach for the system. In this paper, we present a system designed for OST AR headsets that occludes the subject's facial expressions with an emotion-presenting 3D emoji model. We hope this system could help us understand how children with ASD perceive emotions through standard emotion presenting systems and help them enhance their skills of understanding facial expressions.