Dense medium cyclones are widely used separating devices for the washing of plus 0.5mm coal. As approximately fifty percent of Australian coal is washed in these devices, efficient cyclone operation is obviously important. Optimization of dense medium cyclone circuits requires an appreciation and understanding of both cyclone operation and common operating difficulties. Examination of the literature showed that little material has been published addressing such problems. The principal reason for this shortcoming was that the traditional float and sink analysis for the determination of density separator performance was both expensive and time consuming. A survey indicated that the use of density tracers was the most promising alternative technique, although existing density tracers were not made to sufficiently rigorous specific gravity specifications to accurately define coal washing dense medium cyclone partition curves. The statistics of the density tracer process were analysed, resulting in the selection of appropriate i. specific gravity increment sizes, and ii. numbers of tracers per increment to give results of the required accuracy when testing coal washing dense medium cyclone circuits. Appropriate data fitting techniques were developed and tested for accuracy, and estimates of the standard deviation to be expected in tracer-derived partition data were calculated. A development programme resulted in the manufacture and testing of tracers suitable for use at both the operating washery and pilot plant level. The tracers were made to closer specific gravity increments, and more exacting specific gravity, tolerances, than tracers previously used. The density tracers were tested in a number of Australian washeries, and techniques developed for the evaluation of circuit performance both under load and with the feed coal off. Partition data derived from tracer and traditional float sink methods were compared, and the tracer technique was shown to accurately define density separator performance. In the course of the washery testwork, two areas of separation inefficiency were recognized. i. Cyclone operation was unstable and inefficient when the cyclone was fed coal containing a large percentage of near-gravity material, while being operated in such a manner that a large differential existed between the overflow and underflow medium specific gravities. ii. Overall dense medium cyclone circuit efficiency was decreased if the different modules were not separating at identical cutpoints. Cutpoint differences between modules were found to be due to differing degrees of wear on cyclones and pumps, poor specific gravity control, and miscalibration of density gauges. Such problems were found at the majority of plants tested. The subsequent pilot plant testwork examined the effects of common operating variables on separating efficiency. All variables were found to affect either the cutpoint and separation efficiency or the specific gravity range of particle retention by their influence on the ' dynamic medium stability, as measured by the specific gravity differential. Regression analysis was used to develop a medium based model of coarse coal dense medium cyclone separation, and the separation mechanism was explained in terms of this model. Irregular and inefficient cyclone operation was found to be related to the magnitude of the specific gravity differential between the overflow and underflow medium specific gravities and the amount of retention-prone material in the cyclone feed. The retention prone material has a long residence time in the cyclone separating zone, and if sufficient such material accumulates, then the flow pattern in the cyclone is temporarily disrupted. The specific gravity range of retention prone material was shown to increase as the specific gravity differential increased. Efficient separation resulted from cyclone operation with low specific gravity differentials when treating coals containing moderate to high percentages of near gravity material. When treating coals containing little near-gravity material, high specific gravity differentials can be tolerated since the feed contains little retention prone material. Medium segregation was shown to be caused primarily by classification of the magnetite solids. This contrasts with high density ore separations, where medium segregation results from bulk sedimentation of the medium. Areas of further work to fully implement a model of medium segregation are outlined.