The most critical factor affects accuracy of a Structured Light System (SLS) is calibration. Camera calibration is easy to complete because of its extensive study. To simplify projector calibration, previous wor models the projector as an inverse camera and tries to build similar 3D-2D mapping data for projector calibration. Achieved mapping data is directly fed to some classic two-step camera calibration methods. When projector comes with a large distortion lens, this kind of methods will fail because their first steps use closed-form solution to calculate initial guess for optimization in next steps. We proposed a new method to calibrate the projector by removing its distortion first. Because projector cannot “see” anything, not like camera case, constraints such as “straight lines remain straight” working just on 2D image is invalid for distortion estimation. With 3D-2D mapping data, the estimation will involve several extra unknowns into a non-linear optimization. We use partial mapping data whose 2D points in a “small central area” of projector pattern image to acquire an initial guess for those unknowns, and then use all mapping data to refine them and estimate distortion parameters. Experiments show our method can still calibrate the projector when classic methods fail.