Abstract: A software system for pattern recognition of spectrum signal of metal transfer mode was developed using Visual Basic. Based on consistency of the spectrum and metal transfer and geometrical pattern features of the spectrum, several key characteristic parameters were extracted.Correspondent recognition function and a minimum-distance classifier were constructed based on Bayes decision theory. The results show that using this system, pattern recognition of metal transfer mode for metal inert-gas welding, metal activegas welding, and CO₂ welding can be done automatically. This provides a basis for automatic control of metal transfer in metal gas arc welding.