Abstract: As the noise in orbit-derived accelerations is colored,the decorrelation filter based on ARMA(auto regressive moving average) model is proposed to suppress the noise.The principle of the acceleration approach and the least-squares solution using decorrelation algorithm are introduced.And the mathematic model to simultaneously estimate the accelerometer scale factors,bias parameters and geopotential coefficients are also derived.Furthermore,the specific scheme for processing the colored noise of satellite accelerations based on ARMA filter is discussed.The Earth's gravity field model WHUCHAMP-ACC60KP up to degree 60 is recovered from 46-day CHAMP kinematic orbits and accelerometer data.Compared with the models of EGM96,EIGEN-1S,EIGEN-2 and EIGEN-CHAMP03S,the results show that WHUCHAMP-ACC60KP is more accurate than EIGEN-1S,and its total accuracy is near to the model of EIGEN-2,which validates the effectiveness of the proposed method.