The observability analysis method is widely used in performance evaluation of the Kalman filter. Two observability analysis methods, error covariance matrix and Attitude Dilution of Precision (ADOP), are studied in this paper, and the relationship between the them is analyzed for calculating the degree of observability. To the observability of rotating-IMU's multi-position initial alignment, the 33-state error model of the SINS was taken as its study object. Both the two methods are used for research the best rotation sequence, best stated duration and model order reduction in multi-position initial alignment. The simulation of initial alignment observability is carried out, the result of theory research and simulation show the consistency of two methods.