In order to monitor the global terrestrial carbon cycle, NASA (National Aeronautics and Space Administration) provides 8-day GPP images by use of satellite remote-sensing reflectance data from MODIS (Moderate Resolution Imaging Spectroradiometer) at l-km nadir spatial resolution since December, 1999. MODIS GPP algorithm adopts DAO (Data Assimilation Office) meteorological data to calculate daily GPP. By evaluating reliability of DAO data with respect to surface weather station data, we examined the effect of errors from DAO data on MODIS GPP estimation in the Korean Peninsula from 2001 to 2003. Our analyses showed that DAO data underestimated daily average temperature, daily minimum temperature, and daily vapor pressure deficity (VPD), but overestimated daily shortwave radiation during the study period. Each meteorological variable resulted in different spatial patterns of error distribution across the Korean Peninsula. In MODIS GPP estimation, DAO data resulted in overestimation of GPP by $25\%$ for all biome types but up to $40\%$ for forest biomes, the major biome type in the Korean Peninsula. MODIS GPP was more sensitive to errors in solar radiation and VPD than in temperatures. Our results indicate that more reliable gridded meteorological data than DAO data are necessary for satisfactory estimation of MODIS GPP in the Korean Peninsula.