Abstract: In order to achieve hand based natural behavior interaction in augmented reality,this paper proposes a 3D interaction algorithm for marker-free hand.This algorithm uses a series of steps of hand detection,hand feature extraction and hand 3D structure reconstruction to get 3D structure of a bare hand.Based on this,we extract 3 basic 3D interaction modes of hand: point,line and plane in real world and align them in the virtual world to achieve hand based natural behavior 3D interaction in the environment of augmented reality.This paper also presents experimental results to verify the superior performance,accuracy and flexibility of the proposed approaches.