The mechanical properties of an API X65 microalloyed steel were investigated with industrial thermomechanical experiments. The many parameters of processes obtained during production on the plant were systematically changed to optimise the strength and toughness properties. Among these parameters, parameters taking part in secondary metallurgy, such as used collemanite amount for slag formation, or stirring time for reducing of the non-metallic inclusions, which are not viguriously investigated in general, are included in the modelling. The optimised parameters were used for the production of the API X65 steel. However, it is not easy to determine as to what parameters under which conditions influence the toughness properties of the material. Therefore, in this study, a generalised regression neural network was developed to predict the impact energy as a function of experimental conditions. The predicted values of the impact energy using the neural network are found to be in good agreement with the actual values from the experiments. (c) 2005 Elsevier Ltd. All rights reserved.