Abstract: A passive vision sensing system to acquire the image of molten pool inMAG welding was set up and many typical molten pool imageswere obtained.The relationship between the typical gas pores and the characteristics of molten pool images were studied by sync contrast experiments. Many molten pool imageswith surface gas pores and inner gas pores were obtained, and the change of the images and the strange characters were studied from the view of gray mean and gray standard deviation. The experimentsindicate that it is feasible to judge the pores by characteristics of the vision images which offers the technological foundation for automatic identification of welding defect on the base of vision sensing.The experiments and the analysis indicate that one type of image may predicate several types of welding defect, one type of welding defect may also shows several types of image characters.