A method of Cloud Motion Wind (CMW) vector prediction is presented, which is based on the Scale Invariant Feature Transform (SIFT) and achieves the CMW vector results by means of calculating and matching scale invariant feature points. Comparison is conducted between traditional Maximum Cross Correlation (MCC) and SIFTS methods by using the experimental data of cloud pictures from the FY 2 satellite. The comparison result indicates that the SIFT method can achieve more valid cloud vectors under different types of clouds and have more reasonable distribution.