Abstract The design of welding alloys to match the ever advancing properties of newly developed steels is not an easy task. It is traditionally attained by experimental trial and error, modifying compositions and welding conditions until a satisfactory result is discovered. Savings in cost and time might be achieved if the trial process could be minimised. This work outlines the use of an artificial neural network to model the yield and ultimate tensile strengths of weld deposits from their chemical composition, welding conditions and heat treatments. The development of the models is described, as is the confirmation of their metallurgical accuracy.