A robust geometric camera calibration technique with a non-metric planar grid is proposed. For each image of the non-metric grid, a projective model of the grid intersections represented by the grid row and column numbers is derived with sixteen physically meaningful parameters. In this model, intrinsic camera parameters and coefficients of lens distortions are involved. After several images of the same gird have been acquired by the camera awaiting calibrations, two broad steps are taken to work out the camera parameters. Firstly, with two reasonable approximations the projective parameters of all view images are solved linearly and analytically by using the properties of the rotation matrix skillfully. Secondly, with the rigorous projective formulations of non-metric grid intersections, a mathematical optimization model is established to obtain the calibration results with least squares errors. The robustness and accuracy are verified by using many real grabbed images.