People with autism spectrum disorder (ASD) find it difficult to recognize and respond to emotions conveyed by the face. Most existing methodologies to teach people with ASD to recognize expressions use still images, and do not take into account that facial expressions have movement. We propose a new approach that uses state of the art technology to solve the problem and to improve interactivity. It is based on an avatar-user interaction model with real time response, which builds upon the patient-therapist relationship: it is designed to be used by the therapist and the patient. The core technology behind it is based on a technique we have developed for real time facial synthesis of 3D characters.