Vision sensors using optical triangulation have been widely used for automatic welding systems in various ways. Their reliability is, however, seriously influenced by the arc light reflected from the base metal surface. In this study, the reliability of vision sensors was analyzed for the variation of the arc noise by considering the reflectance of the base metal surface. The property of the surface reflection of the base metal was modeled using the bidirectional reflectance-distribution function (BRDF), and then the intensity variation of the reflected arc was formulated for various configurations of the torch, base metal and sensor. The experimental data of the arc light reflection were obtained for two materials, mild steel and stainless steel, each having different surface reflection characteristics. It was found that the results calculated from the proposed model were in good agreement with the experimental data.