In this paper, we present an approach for smooth surface reconstructions interpolating triangular meshes with arbitrary topology and geometry. The approach is based on the well-known radial basis functions (RBFs) and the constructed surfaces are generalized thin-plate spline surfaces. Our algorithm first defines a pair of offset points for each vertex of a given mesh to enhance the controllability of local geometry and to assure stability of the construction. A linear system is then solved by LU decomposition and the implicit governing equation of interpolating surface is obtained. The constructed surfaces finally are visualized by a Marching Cubes based polygonizer. The approach provides a robust and efficient solution for smooth surface reconstruction from various 3D meshes.