Abstract: The efficiency of welding defect intelligent recognition algorithms is restricted by many conditions such as image quality and its own adaptability,which results in a low accuracy of welding quality classification.A novel rough set based algorithm of welding quality decision was proposed to improve the accuracy of welding quality classification.Firstly,the rough characteristic of the welding defect intelligent recognition was analyzed.Secondly,the rough set model of intelligent classification was founded by taking classification basis as condition attributes and quality levels as decision attributes.And then, several groups of typical samples were chosen from the defect library.The knowledge base was founded to form the decision table.Lastly,the minimum decision rules were obtained after calculating and reducing the redundant information.The experimental result indicates that the accuracy increase from less than 80% to 94% of that of the ordinary algorithms.Furthermore,it is robust to the migration errors during the recognition process.