Abstract: The problems encountered when using Kalman filter for satellite precise orbit determination are discussed.And a new nonlinear adaptively robust filter is proposed.Firstly,the new method uses nonlinear filter to improve the orbit's precise,avoiding the bad influence of linearization error to orbit results.Secondly,bi-factor variance expanding model is used to adaptively adjust the covariance matrix of measurement noise,and resist the bad influence of measurement outliers to orbit results.Thirdly,an adaptive factor is used to adjust the covariance matrix of state noise,and reduce the bad influence of state model errors to orbit results.An example of CHAMP satellite orbit determination is given,and the result demonstrates that the new method is feasible and effective.