Abstract A large data compilation has been assembled of platinum group element (PGE) analyses in mantle melts and mantle rocks, the latter including an assortment of xenoliths and obducted mantle massifs. The degree of correlation has been investigated among the PGEs and with other major element variables such as Al2O3, TiO2 and Mg number, and the results are considered in the context of the current paradigm for the behaviour of highly siderophile elements in the silicate Earth. Primitive mantle melts have a wide range of PGE contents. Komatiites have the highest abundances of all the PGEs, show the strongest correlations between Pt and Rh, Pt and Pd and between the iridium-group PGEs Ir, Ru and Os (IPGEs). Most basalts of all affinities have lower levels of Pt and Pd and much lower levels of Ir, Ru and Os than komatiites. Within the basalt grouping Rh has stronger affinities with the IPGEs. Picrites and Archaean basalts are intermediate between these two groups. MORBs and a small proportion of continental LIP basalts show strong depletions in all PGEs attributable to retention of sulfide in their mantle source rocks, or sulfide liquid fractionation on ascent. The degree of PGE depletion in other basalts is probably attributable to equilibration with sulfide, but is less than would be expected under conventional models of sulfide extraction, and is instead attributed to mixing of magmas generated at variable depths incorporating both sulfide-saturated and undersaturated components. Basalts with Pt and Pd contents higher than typical komatiites are rare, a notable example being B1-type parent magmas to the Bushveld Complex, which have komatiite-like relative PGE abundances and Pt, Pd and Rh abundances up to a factor of two higher than komatiites for comparable Ti contents. The mantle composition array as a whole is characterized by variable degrees of depletion of Pt, Pd and Rh in Al-poor, melt-depleted harzburgite/dunite lithologies; lack of depletion in these elements in Al-bearing lherzolites; and a lack of systematic variation in IPGEs across this range. Strongest correlations across the entire set are observed between Ir, Ru and Os; and between Pt and Rh. Melt-depleted cratonic mantle samples are notably more deficient in Pd than in Pt, but comparable Pd-enriched components are not represented in the available data from continental environments. The only group of mantle melts that systematically record high Pd/Pt ratios are MORBs; if these are indeed the complement of the depleted cratonic mantle suite then the melt depletion recorded by the cratonic mantle suite occurred at low pressure prior to tectonic underplating of the depleted lithosphere beneath the cratons. A filtered subset of orogenic peridotite compositions that are thought not to have been affected by significant extents of melt extraction or metasomatic refertilization have median concentrations of 3.9 ppb Os, 2.9 ppb Ir, 6.3 ppb Ru, 1.0 ppb Rh, 6.2 ppb Pt, 5.4 ppb Pd, and Cu/Pd ratio of 5500, which we consider to be representative of the modern convecting mantle. The convecting mantle has PGE proportions closely resembling those of lunar impact breccias, diverging considerably from chondritic proportions and attributable to the presence of a late veneer-derived, predominantly sulfide-hosted component. The compositions of mantle peridotites show considerable scatter attributable to the combined effects of measurement error and a strong covariance due to a heterogeneous distribution of sulfide in the small samples typically chosen for pulverization. The intensity of the covariance between all of the PGE due to sampling error gives a false impression of a genetic trend toward highly enriched PGE in some samples which could be mistaken for the effects of metasomatism; however no plausible metasomatic process would be expected to retain the tight interelement correlations shown.