A closed-loop linear covariance analysis method was proposed for orbital rendezvous using AON (angles-only navigation). The SRUKF (square root unscented Kalman filter) based on AON algorithm was constructed and the observation sensitivity matrix was further calculated. The multi-impulsive Hill guidance law was employed to derive the closed-loop linear covariance analysis model. The results of the numerical simulation indicate that the closedloop linear covariance analysis result fits the 1000 times Monte Carlo shooting well. The covariance analysis method is applicable to the traditional EKF (extended Kalman filter) based on the AON method, but has an estimation bias along downrange, which is equivalent to the variance of trajectory dispersion. The major axis and minor axis of error ellipse achieved with EKF based on covariance respectively are about 24.68% and 20.56% longer than the results from SRUKF based error ellipse. Besides, SRUKF and EKF have the same order computational burden for the state estimation, but the SRUKF is about 10% faster than the EKF due to using two powerful linear algebra techniques, QR decomposition and Cholesky factor updating.