Dense medium drum separator efficiency was analysed at two Western Australian iron ore concentrators. Efficiency data had hitherto been difficult to accumulate due to the high cost of analysing representative, and therefore large, coarse ore samples. To overcome this problem, recoverable density tracers, which are in effect artificial mineral and gangue, were manufactured in three different sizes (15 mm, 30 mm and 60 mm). Their use enabled partition curves to be studied within one hour of conducting a test: separation densities were generally lower than the feed medium density for all three sizes of density tracer. For a given set of operating conditions, the smaller tracers had lower separation densities and larger ecart probab/es than the large tracers. These phenomena are explained with the aid of a conceptual model of a drum separator, which assumes that the feed ore plunges to the bottom of the drum upon introduction. Particle rising velocities are, therefore, process determining.Medium rheology was altered by adjusting the proportion of solids in the pulp. This was accomplished by changing both the pulp density and ferrosilicon/magnetite ratio. Higher solids volume fractions invariably caused a decrease in separation densities and an increase in ecart probables, most markedly for the smaller tracers. A similar response was observed when feed tonnages were raised.An explanation for these phenomena is sought in the discussion which also suggests means of improving drum separator efficiency.