Abstract: With the rapid development of VR industry,the non-contact gesture interactive application is increasingly common. In this paper,a modeling method with Leap Motion was proposed for rapid 3D VR scene developing of ancient architecture. Firstly,hand-motion data were captured with Leap Motion to implement the real time mapping,interactive controlling and editing of physical and 3D space. Secondly,a probabilistic model based recommending method was proposed for building components of ancient architecture. By designing a Bayesian network,the relationships between different components were calculated by probabilistic inference. Finally,several functions were implemented for rapid developing of ancient architecture scene,including returning the recommending results to user in real time,choosing corresponding components by gesture and optimizing the positions in the 3D space directly. Experimental results show that the method can take use of the 3D space information more effectively than traditional 2D style,and much more intuitive,convenient and efficient.