Samples of the feed, underflow and overflow mediums from dense medium cyclones operating at low separation densities in eight commercial coal washeries were sized and analysed, to obtain comparative cyclone classification performance curves for the magnetite, shale and coal components in the circulating medium. The resulting component classification curves indicated that the dense medium cyclones hydraulically classify the magnetite and the shale to the underflow with a Stokesian size and density relationship at these low medium densities, ie the shale curve was shifted parallel to and appropriately coarser than the magnetite curve because of the lower density of the shale. However the lower density coal reverse classified to the overflow (ie away from the cyclone wall) with a similar Stokesian relationship. By correcting the magnetite and shale classification curves for the bypass inefficiency to underflow, and similarly correcting the coal classification curves for the bypass inefficiency to overflow, the corrected classification curves for the magnetite, shale and coal were reasonably parallel (considering that the coal has a range of densities). Using a value of five for the density of the magnetite, the displacement of the shale and coal curves corresponded to densities of 2.27 and 1.38 respectively for one circuit, although the data was not as good for the other circuits. This reverse classification mechanism was confirmed in a 140 mm diameter laboratory dense medium cyclone circuit using low density (1.20 - 1.30 RD) tracers, sized from 0.25 to 2 mm. These classification performance curves were corrected for bypass ineffiency (as above). The corrected curves were parallel and displaced according to tracer density, even though the lowest density tracers were reverse classified to the overflow. The data indicate that the rates of classification of all particles could be described by the same Stokesian relationship, regardless of the direction of travel, except for a narrow band at the partition density. Particles heavier than the partition density classified to the underflow, particles lighter than the partition density reverse classified to the overflow and particles identical to the partition density were not classified at all. Consequently it should be possible to simulate partition in dense medium cyclones, operating under free settling low density conditions, with a simple cyclone classification model, provided the partition density can be accurately calculated from, or correlated with the cyclone feed, underflow or overflow pulp densities. The tracer partition performance in the laboratory cyclone was an inverse function of the tracer particle size, as proposed by King and Juckes (1984) for medium/coal ratio in low density medium systems. Their sized coal data was parallel, with lower partition values due to a higher energy cyclone. Partition data with similarly sized tracers (Scott, 1988) showed a parallel shift to higher partition values due to the separate effects of higher viscosity medium and a larger diameter lower energy Cyclone.