The separation selectivity and efficiency of the flotation process depends not only on differences in the physico-chemical surface properties of various minerals within an ore, but also on the hydrodynamics of flotation. The hydrodynamics relate directly to many sub-processes of the flotation, such as suspension of particles in the pulp, dispersion of the air into bubbles and the probability of particle-bubble collision. Mechanical entrainment plays a major role in the quality of the flotation concentrate and the separation selectivity. Therefore, the effects of some important hydrodynamic parameters on both the entrainment and the flotation performance were investigated by means of a fractional factorial experimental design. Furthermore, many flotation tests were performed in order to generate sufficient data for use in empirical modelling and neural network based modelling, in light of the results of the experimental design. The adaptability and reliability of the models developed, depending on the treatment of available experimental data, have been studied comprehensively. Evaluation of the observed and predicted results demonstrated that the effect of some chemical and hydrodynamic parameters of the flotation process on both the metallurgical performance and entrainment in the training region, can successfully be predicted, with an error of less than 6%, by using the developed neural networks models without particular assumptions and additional experiments. In order to prove the validation of the empirical and neural networks models developed and to compare their performances, a few additional tests were conducted under predetermined flotation conditions. The results of these additional flotation tests indicated that the neural network models were consistently more accurate than the empirical models with a negligible error. (C) 2003 Elsevier Ltd. All rights reserved.