A series of test simulations are performed to evaluate the impact of satellite-derived meteorological data on numerical typhoon track prediction. Geostationary meteorological satellite (GMS-5) and NOAA's TIROS operational vertical sounder (TOVS) observations are used in the experiments. A three-dimensional variational (3D-Var) assimilation scheme is developed to assimilate the satellite data directly into the Penn State-NCAR nonhydrostatic meteorological model (MM5). Three-dimensional objective analysis fields based on the T213 results and conventional observations are employed as the background fields of the initialization. The comparisons of the simulated typhoon tracks are carried out, which correspond respectively to assimilate different kinds of satellite data. It is found that, compared with the experiment without satellite data assimilation, the 3D-Var assimilation schemes lead to significant improvements on typhoon track prediction. Track errors reduce from approximately 25% at 24 h to approximately 30% at 48 h for 3D-Var assimilation experiments.