GF-4 is the first Chinese high resolution geostationary orbit satellite. It has great application potential in many earth-related studies. Given the low geometric accuracy of GF-4 images in mountain area, in this paper, a new operational and practical automatic orthorectification approach was proposed to improve the orthorectification accuracy of GF-4 images. The new approach adopted a two-level area-based algorithm to automatically search tie points between GF-4 and the base Landsat images. Then the images was further orthorectifyed using the improved rational polynomial coefficients model optimized by tie points. Results demonstrated that the new orthorectification approach could significantly improve the orthorectification accuracy. It is also suitable for orthorectification of GF-4 images with different clouds coverage.