Emerging applications of active 3D technology have raised expectations on development of more user-friendly stereoscopic 3D content. To optimize stereoscopic 3D content by taking into account human factor, understanding how user gazes at 3D direction in virtual space is an important research topic. In this paper, we suggest a novel approach to estimate 3D point of gaze on stereoscopic display using optimized geometric method. Dual-camera system and Direct Linear Transformation (DLT) algorithm have been utilized to compute user dependent parameters and 3D coordinate of pupil. Instead of asking user to perform a time-consuming 3D calibration session using more than twenty calibration points, a simplified 3D calibration method using only three calibration points is proposed. Our experimental results show that the proposed method achieves better accuracy compared with conventional geometric method by average errors 0.83, 0.87, and 1.06 cm in X, Y, and Z dimension, respectively. Compared with subjective depth judgment, the proposed gaze tracking system is more robust in measuring depth of virtual 3D object with various sizes.