The challenge to predict variations in penetration depth in one-sided fillet welds during robotized gas metal arc welding has been addressed by a pilot investigation of technical possibilities and limitations. The main cause for the variation in penetration depth is considered to be variation in joint gap size. Special attention has been paid in order to adopt the experimental conditions to conform to industrial welding conditions. The employed method uses in-process monitoring of joint gap size together with an empirical model relating penetration depth to gap size in order to predict this depth. The gap size estimates are based on image information from two cameras, one visual and one infrared. The results, that are evaluated off-line, confirm the development of a real-time method providing technical solutions that are industrially tractable. The results also pinpoint areas of further improvements towards increased robustness and reduced estimation uncertainties.