Because of the difficulty in applying arc sensors to the welding of thin corrugated plates, vision sensors that provide three-dimensional geometry information by optical triangulation have been widely used for robot guidance in automatic welding of plates. The reliability of vision sensors, however, is influenced considerably by arc light reflected from the base metal surface in spite of blocking the direct arc light with a light isolator. A reflectance model of a welding arc was developed to estimate arc noise for various configurations of sensors, base metals, and welding arcs in three-dimensional space by assuming the welding arc as a point and extended light source. Various experiments were conducted to determine the bidirectional reflectance distribution function (BRDF) parameters of the model, and to verify the validity of the proposed model. Two proposed models, the point source model and the extended source model, were compared with, the gray level of the reflected arc caught by a CCD camera. The experimental data of the gray level of the reflected arc generally agreed well with. calculated results obtained by the two models, while the model with a point light source resulted in a greater discrepancy for short distances between the welding arc and reflecting surface of the base metal. The proposed models and experimental results can be effectively used for optimal design of the configuration and the moving path of vision sensors according to base metal shape, which improves the measuring efficiency of vision sensors (Refs. 1, 2) and reduces the effect of arc noise on them (Ref. 3), thus enhancing the performance of the vision sensor in automatic joint tracking of the arc welding process.