The proposed method is a robotic system for behavior intervention of children with autism who have deficit in development of communication and interaction with others. This robotic system helps the social training of children with autism in that robot reacts as a social being. We aim to improve the eye contact skill of children with autism, one of main indicators of their development. This paper describes our hardware configuration and methods for analyzing a child behavior based on vision and auditory information, and our human-robot interaction (HRI) architecture. At the end, preliminary experimental results with a non-autistic child will be presented.