The quality and error distribution of atmospheric motion vectors (AMVs) derived from the water vapor band of FY-2C meteorological satelite of China in 2009 were analysed. The results indicate that the errors of AMVs varies inhomogeneously in different climate zones, seasons, and vertical positions in space. Based on the thermal wind theory, we noticed that over estimation in the value of height may be the dominant origin of the AMVs errors. We verified the conjecture by reassigning the AMVs to new heights, with the NCEP/FNL reanalysis wind field as a reference. The quality of AMVs before and after reassignment was compared with the NCEP/FNL data and radiosonde observations of international exchange stations. It’s found that the negative biases of the U-winds of AMVs decrease explicitly from -9.73 ms-1 to -1.1 ms-1, the standard deviations of errors from 10.24 ms-1 to 4.5 ms-1, the quality of which is improved over 50%. The V-winds are also improved obviously. The seasonal errors of AMVs are removed well. All of the improvements justify the aforementioned viewpoints. This paper suggests a preliminary method which assigns the AMVs to reasonable heights with the reanalysed wind field of a NWP model as a reference, and also supplies an idea of detecting algorithm for AMVs producers.