This paper presents a robust approach based on evolutionary agents for projective reconstruction in the presence of missing data and unknown depths. Agents denote possible submatrices for rank constraints, and carry out some evolutionary behavior to exploit a vast solution space. Our approach combines the benefits of excellent searching ability of evolutionary agents for getting a good solution, with a proper treatment of missing information with linear fitting. Experimental results demonstrate better performance of our approach than other typical methods in terms of accuracy and robustness to noise and missing data.