The Ausmelt(R) converter is being introduced into the South African platinum industry as an alternative to the traditional Pierce Smith converter process to convert a nickel-copper matte containing platinum group metals (PGMs). The semi-fundamental modelling technique used involved developing a complete thermochemical equilibrium database of the system, and then using this in conjunction with linear systems models to predict the dynamic iron and sulphur concentrations in the matte, The thermochemical equilibrium database was developed using the FACT thermochemical simulation software. The generated equilibrium database was used to train and validate feedforward neural networks. This technique allowed the database to be used online to predict the matte equilibrium composition. The equilibrium matte concentrations of iron and sulphur predicted by the neural network were then used as exogenous inputs into two linear systems models used to predict the actual iron and sulphur matte concentrations, It was found that the models characterised the dynamic behaviour of the sulphur and iron well. (C) 2002 Elsevier Science Ltd. All rights reserved.