Abstract: In view of the problem that the noise of the inertial-stellar attitude determination system is non-Gaussian distribution,selected unscented particle filter(UPF)for the attitude determination system.UPF uses the unscented Kalman Filter(UKF)to generate sophisticated proposal distributions.It can avoid the limitation of the Extended Kalman Filter(EKF)and the UKF which only apply to Gaussian distributions.This paper takes the Micro Electro Mechanical System(MEMS) gyros and CMOS APS(Complementary Mental Oxide Semiconductor,Active Pixel Sensor)star sensor as attitude sensors,the method of attitude determination uses minimal-parameter attitude matrix estimation as the attitude parameter,presents a UPF attitude estimator.The attitude determination filter was constructed and semiphysical simulation was adopted from the MEMS gyro,and the unscented particle filter's characters are analyzed and compared with the EKF and the UKF.The simulation results show the system performance is improved by the suggested method.The convergence rate and the determination accuracy are higher than the EKF and the UKF.