Abstract: A PAW (plasma arc welding) longitudinal seam tracking system was introduced,which senses the molten pool and the seam in the same image by a vision sensor,then calculates the seam deviation and adjusts the work piece adaptively move to the seam position sensed by the vision sensor.A novel molten pool area image-processing algorithm based on machine vision was described.The algorithm processes each image at the speed of 20 frames per second in real-time to extract three feature variables then which are used to calculate the seam deviation.It is proved experimentally that the algorithm is fast and effective which can satisfy the requirements of the system's real-time response ability.