The base of separation at Gol-e-Gohar iron ore plant (Sirjan in the south of Iran) is magnetic separation after milling. Magnetic separation in the plant is divided into two parts: dry magnetic separation and wet magnetic separation. Sixty-five per cent of the plant production belongs to the dry magnetic separator section. Therefore, investigation of this part of the plant is necessary. Prediction of plant production is one of the most important subjects in mineral processing plants. That means, before feed enters the plant, specification of production has been calculated with laboratory methods. The purpose of this paper is the prediction concentrate quality has achieved from the dry magnetic separator. The content of iron in the final dry concentrate will be predicted, with laboratory, pilot and modelling methods, before feed enters the plant. Different variables are used in this approach, which consist of: speed of the first dry magnetic separator (rougher), speed of the third dry magnetic separator (cleaner), iron content in plant feed, size of 80 per cent passing of feed, size of 20 per cent passing of feed and iron content in pilot dry magnetic separation concentrate that performs in a standard manner. The artificial neural network is used for modelling and one of its branches is the radial basis network (RBN). Iron content in magnetic separating concentrate in plant will be introduced as the target and will be calculated with the RBN method. In the RBN method, a set of data enters to neural network and they will be trained and simulated. Finally, the target will be predicted. Ultimately, the original target (actual iron content) and the predicted target (predicted iron content) compare and error of them will be calculated. The obtained result is acceptable and satisfactory.