In this paper, a NURBS approach based on MPEG- 4 standard for facial modeling is presented. In this method, human face is simulated by using a piece of NURBS surface, and the model points of NURBS surface are affected by MPEG-4 feature points. The method combines the superiorities of NURBS and MPEG-4. For individual object, by means of capture of the MPEG-4 feature points and a mapping relationship, individual facial model is constructed. In some non-continuous region such as eyes, a partition algorithm of NURBS surface is implemented and some necessary aided models such as eyeballs are constructed. Due to the low data quantity of MPEG-4 and the local support property of NURBS, the model can be extended to dynamic modeling of facial expressions. The result of simulation is satisfying.