Automation of arc welding process is very important to improve weld quality and cost performance. However, arc welding process has not fully understood, because of less knowledge on arc bihavior. The welding process was captured by high-speed video systems to analyze dynamic behavior of molten pool Some surface defects such as under-cut, humping bead, bum through. etc can be detected. Internal defects such as blow holes, cracks, lack of fusion, etc are very difficult to estimate their existences from surface observation. However, dynamic behavior of surface condition changes by their existences. So, it is possible to predict defects by image analysis of surface movie images.