A profile of temperature and relative humidity retrieved from Mt. Fuji observed GPS “Downward Looking (DL)” data was assimilated into mesoscale weather prediction model by using four-dimensional variational data assimilation (4D-var) procedure for typhoon case of September 9, 2001. The DL observation offered the profile of the atmosphere over the ocean where typhoon approached. Because the retrieved case was few, the observation error was expediently decided as 1 centigrade for temperature and 4 percent for relative humidity without a statistical investigation. The assimilation results show a small but positive impact for precipitation forecast. But the position of the typhoon center in the initial field slightly shifted to the opposite direction from the best track analysis by the Japan Meteorological Agency (JMA). To decide observation error of DL retrieved refractive index profile, error estimation using a three-dimensional (3D) ray-tracing model which uses mesoscale weather model outputs was executed. The 3D ray-tracing model simulated propagation of GPS signal in the model atmosphere every one-second. Then, Doppler shift, bending angle, partial bending angle (PBA), and finally refractive index profile were retrieved. It was proven that PBAs are able to reproduce from Doppler shift in high accuracy. Error of retrieved refractive index showed high correlation with horizontal variation of refractive index. The results suggest that we should assimilate bending angle or excess phase delay rather than profile of retrieved refractive index, temperature and humidity.