Abstract: Radial basis function （RBF） networks,combined with implicit polynomials,can be employed to represent 3D surface from 3D unstructured points,which are constructed from the zero-set of the RBF networks.The algorithms aim to use the capability of interpolation and fitting of RBF to construct 3D surfaces from neural networks by selecting the exterior and interior constraint points simultaneously.Simulation results show that the algorithms are more robust and stable than the algorithms based on BP networks for small scale of points,and have better fitting results for a few 3D unstructured points than the algorithms based on BP networks which can only get unclosed figures or have many spurious zero-sets.