Accurately calibrating light field camera is essential to its applications. Rapid progress has been made in this area in the past decades. In this paper, detailed analysis was first performed towards the state of the art projection models for calibration which were further interpreted in three representations, including the correspondence between rays and pixels, 3D physical points and pixels and between 3D physical points and 3D signal structure of the captured light field. Based on the analysis, parameters in the projection model were grouped into direction parameter set and depth parameter set. A two-step calibration method was then proposed with each step dealing with each set of parameters. The proposed method is able to reuse traditional camera calibration methods for the direction parameter set. A simply raw image-based calibration of depth parameter set was further proposed. Systematic validations were conducted to evaluate the performance of the proposed calibration method. Experimental results show that the accuracy and robustness of the proposed method outperforms its counterparts under various benchmark criteria.