Abstract: n the acceleration approach for recovery of the Earth’s gravity field, the observation of satellite accelerations are derived from the satellite orbit data by numerical differentiation. However, the differentiation will rapidly amplify the high-frequency noise, which will degrade the accuracy of recovered gravity field model. Because the noise in the orbit-derived acceleration data is colored, whitening filters based on decorrelation technique are introduced to suppress the noise. Two whitening filters are constructed based on 3-points differential scheme and on ARMA model, respectively. As a test, simulated CHAMP orbit data with different type of noises are used to recover the gravity field model. The results demonstrate that the gravity field models recovered from the decorrelation filtering methods have higher accuracy than those from equal weight method, which validate the effectiveness of the decorrelation filtering methods.