A computational fluid dynamics (CFD) model of the dense medium cyclone (DMC) has been developed, using Fluent, by coupling component models for the air-core, the magnetite medium and coal particles. Simulations of turbulent driven flow in a dense medium cyclone with magnetite medium showed that the predicted air-core shape and diameter were close to experimental results measured by gamma ray tomography. Multiphase simulations (air/water/medium) using the large Eddy simulation (LES) turbulence model, together with viscosity corrections according to the feed particle loading factor, gave accurate predictions of axial magnetite segregation, with results close to gamma ray tomography data. Addition of lift forces and viscosity correction improved the radial magnetite segregation predictions especially near the wall. Predicted density profiles are very close to gamma ray tomography data, showing a clear density drop near the wall. At higher feed densities the agreement between the empirical correlations of [Dungilson, M.E., 1998. A model to predict the performance of the dense medium cyclone for low and high density applications, In: Seventh JKMRC Conference, Brisbane, Australia, 67-84; Wood, J.C., 1990. A performance model for coal-washing dense medium cyclones, Ph.D. Thesis, JKMRC, University of Queensland] and the CFD are reasonably good, but the overflow density from CFD is lower than the empirical model predictions and experimental values. It is believed that excessive underflow volumetric flow rates are responsible for under prediction of the overflow density. The partition characteristics of the DMC for particles between 0.5 and 8 mm in diameter were modeled using Lagrangian particle tracking. For the first time, the pivot phenomenon, in which partition curves for different sizes of coal pass through a common pivot point, has been successfully modeled using CFD. The values of E-p predicted by the Lagrangian particle tracking are very close to the experimental values although cut-point predictions deviate slightly. This comprehensive CFD model provides a tool for new DMC design with clear advantages over approaches based on constructing and trialling new designs experimentally. (c) 2006 Elsevier Ltd. All rights reserved.