A novel visual robotic arc welding system based on circular laser vision sensor is developed. After image de-noising, image segmentation, and image thinning, the relation of depth value of workpiece and off-axis angle 'gamma', three-dimensional (3D) calculation, and seam tracking experiments are carried out. Finally, the error for seam tracking system is analyzed. The results show that 1) 3D information can be obtained using the proposed visual robotic arc welding system and the real-time seam tracking is realized; 2) the seam tracking error is small enough for gas tungsten arc welding (GTAW) process, and this system can be used for seam location and seam tracking or seam finder.