Because of the disadvantage of the existing scheme of variational correcting for remote sensing data, an improved scheme was brought up. In the new scheme, it is not necessary that the correction reference value would be interpolated to the grid of remote sensing data. Therefore, the difficulty of objective analysis and the bad impact of interpolation errors can be avoided in the region of reference value lack. By means of the scheme, the errors of remote sensing data can be reduced evidently while its spatial pattern is held. Some tests were performed to correct the ATOVS retrieved data from the National Satellite Meteorological Center, and the results show that the improved scheme is better in correcting remote sensing data. The ATOVS retrieved data corrected are put to use in Grapes-3Dvar system, and have proved applicable by the application experiment.