The purpose of the investigation was to develop a procedure for optimisation of complex flotation circuits, using steam assays and pulp densities alone (i.e. no additional measurements or batch tests). Conventional models, based on pulp kinetics, were regressed directly to steady-state data from a nine-stage platinum flotation plant, to provide a basis for optimisation studies. 'Froth factors' were obtained by regression to characterise the froth in each stage and a 'depressant factor' was also required for gangue minerals, to account for effect of additional depressant in the cleaners. These factors improved the fit, but significant differences between plant data and the model were still present, particularly in the cleaning circuit. The model for the froth was modified, to incorporate two rate constants for the recycle of particles from the froth. These parameters provided a workable approximation of the distribution of probabilities of particles being recycled in plant cells, depending upon where particles appeared at the surface. In order to maintain an acceptable number of model parameters, only two pulp kinetic parameters and two recycle parameters were used for the three mineral classes. The fit to the data was improved very significantly, when compared to the use of conventional pulp-based models with several pulp rate parameters. This relatively simple model did not require additional data from batch tests and it was suitable for plant simulations. Simulations of various strategies for plant operation were tested, allowing the simulator to vary the concentrate flows from all or some of the stages, in order to maximise platinum recovery. The simulator was imbedded in an automatic optimiser and constraints on chrome content or mass flow of final concentrate were used. The simulations of the existing circuit demonstrated that an improvement in recovery of about 0.5 per cent could be achieved by optimising all concentrate flows. The mass pull of concentrate per unit area was used as a guide for maintaining efficient cell operation and in some cases cleaner cells were re-allocated to modify the volume of a stage. The addition of an extra cleaning stage was simulated, resulting in a further increase in recovery. Plant tests are in progress to verify the simulations. (C) 2005 Elsevier Ltd. All rights reserved.