Abstract: A passive vision sensing system for taking the image of MAG welding pool had been set up.CCD and the near-infrared compound filters system were used to eliminate the arc disturbance,by which clear images were obtained.The relationship between image characteristic information and the weld defect was studied by sync contrast experiments.The visual characters of the pore,weld collapse and the other defect were extracted.Experimental results show that it is feasible to judge the defect by visual characters which offers a technological foundation for automatic identification of welding defect on base of vision sensing.