Abstract: The near space global atmospheric temperature field from 20-100 km was achieved using 3-dimensional variational (3DVAR) assimilation method, of which the observation data was taken from TIMED\SABER temperature data and the background data was taken from WACCM model. Obvious variations could be seen in the near space global atmospheric temperature field after 3DVAR assimilation. An evaluation analysis based on statistical method was accomplished. The results indicate that the errors of the near space global atmospheric temperature field get a general decrease after 3DVAR assimilation, with the maximum error decreasing from 17 K to 7 K. The application of this 3DVAR assimilation algorithm can provide more accurate initial fields to near space atmospheric environment forecast model.