Abstract: When autonomous optical navigation is used to achieve accurate planetary landing, the amount of computation required for landmark selection needs to be reduced due to the limitation of computational resources. In this paper, a depth estimation error-based landmark selection method was proposed. First, a depth estimation error model was developed to describe the accuracy of distance estimation between the lander and the landmark when the same landmark was observed twice in a row, and then the model was used to describe observability degree and to select the landmark with highest accuracy of line-of-sight depth estimation from the sequential images. The landmark selection method proposed in this paper is similar to the conventional methods in terms of navigation accuracy, but requires less optimization computation time and is more suitable for autonomous navigation on the lander.