Abstract: The Sparse Recovery Space Time Adaptive Processing (SR-STAP) algorithm can effectively improve the clutter suppression ability of airborne radar in complex environment. Usually, the space-time plane is uniformly dispersed into several grids to construct the dictionary. However, the real clutter points often cannot fall on the pre-discrete grid points, and the off-grid effect will occur at this time, leading to the performance degradation of SR-STAP. To solve this problem, this paper proposes a knowledge-assisted dictionary correction method. Firstly, a priori knowledge such as the parameters of the carrier platform is used to uniformly discretize the spatial frequencies. Then, the Doppler frequencies are calculated and corrected, and the spatial frequencies are corrected according to the priori knowledge. Finally, the space-time steering vector corresponding to the revised spatial frequencies and Doppler frequencies is used to construct a super-complete sparse dictionary. The simulation results show that, compared with the traditional dictionary construction algorithm, the dictionary correction method can effectively overcome the off-grid effect and improve the performance of STAP.