Abstract: The azimuth was often a large angle because navigation accuracy of fiber-optic inertial navigation system was not high, and the filtering equations of initial alignment were non-linear. In order to improve the initial alignment accuracy of nonlinear models, a improved Sage_Husa adaptive kalman filtering method was put forward, and applied to initial alignment of fiber-optic inertial navigation system. Established initial alignment nonlinear model of large azimuth misalignment angle, contributed the system noise statistics estimators, and used improved Sage_Husa adaptive kalman filtering to simulate for nonlinear error equations. The simulation results show that the improved Sage_Husa adaptive kalman filtering could deal with nonlinear problems, improve the accuracy of initial alignment. Azimuth misalignment angle error estimation precision improved 27% than EKF.