A review of the literature showed that although at present there are several medium based models available for D.M. cyclones, none can claim applicability to both ore and coal as the development has been separately carried out in these areas. The initial objective of the present work was therefore, to synthesize various media based D.M. cyclone models into a single model which is applicable to both ore and coal, and which includes particle size effects. It was also proposed that the new model be validated at industrial site(s). The high cost of generating size by size density distribution data (for surveys done at the Tom Price concentrator), led to an evaluation of a new numerical method proposed by Lewis and Tucker (1986) for generation of density distribution data. The method has the advantage that it can handle all size and density ranges and clearly minimizes the use of toxic heavy liquids and hence, is cost effective. The synthesis of various media based D.M. cyclone models was restricted to Scott's (1986) ore-based model and Wood's (1990) coal-based model. Scott's model based on a pivot phenomenon was applied to Wood's industrial size by size data on coal separation. In addition, Wood's model was applied to Scott's size by size data on ore separation. Since both the Scott and Wood model are medium based, a model to predict medium stream densities is essential. In the present work the following four models were evaluated: a) Scott model - (an empirical model) b) Nageswar Rao model - (based on classification) c) Holland Batt model and - (based on bulk sedimentation) d) Dimensionless model - (derived from dimensional analysis) Scott's pivot phenomenon model structure was found to apply in coal separation while Wood's model could not be applied to Scott's ore data without further modification. The medium stream densities were found to be best fitted using Scott's model, although the Dimensionless model showed promise. Industrial surveys were undertaken at Hamersley Iron's Tom Price concentrator for model validation. Due to high cost of generating size by size density distribution data it was proposed that only two tests be fully analyzed (size by size) using heavy liquids and the remaining six tests be partly analyzed using heavy liquids, for which the full density distribution data would be generated using the modified Lewis and Tucker method. However, because of circumstances beyond the author's control, only the two tests which had been fully analyzed using heavy liquids were finally available for validation of the synthesized model. It is concluded that a medium based model is now available which can be applied to both ore and coal, which includes size effects. A modified version of the Lewis and Tucker method is proposed for generation of density distribution data. The modified Lewis and Tucker method needs an accurate knowledge of the assay grid for successful application of the method. At the end of the thesis suggestions have been outlined for further areas of work to improve the accuracy of Scott's model and to further refine the modified Lewis and Tucker method to broaden its area of application.