Comprehensive numerical simulations of particle laden gas flows are complex in general and its complexity increases dramatically when real world equipment are modelled. In this study, several model parameters required to simulate such a flow are tested. Initially, single-phase numerical results from a static classifier are validated against published experimental data. The Reynolds Stress Turbulence Model is demonstrated to be capable of accurate predictions for classifier airflow. The model is further extended by the addition of a particle Lagrangian phase. Several particle parameters are investigated and compared with published experimental data. It is found that the particle separation accuracy is improved with the inclusion of turbulence dispersion. Particle rough wall model and wall restitution coefficients influence the flow of coarse particles alone. The model is extended to simulate gas flow in a dynamic classifier with rotor. Physics involved in the flow inside a dynamic separator are explained. Copyright © 2019 Inderscience Enterprises Ltd.