Gold is a strongly chalcophile element and its concentration and behaviour in the mantle is intimately connected to the nature, abundance and mobility of sulfide phases. In this study Au has been analysed together with a suite of chalcophile and siderophile elements in 453 sulfide grains hosted in a variety of mantle-derived peridotite and pyroxenite xenoliths from localities in Australia, China and Norway. These data have been combined with analysis of silicate phases to assess the mobility of Au, and sulfides in general, during mantle processes. The Au content of peridotite-hosted sulfides ranges from <0.001 to 40.9 ppm Au. Large variations are observed between sulfide grains within single xenoliths, between xenoliths and between xenolith suites in whole-rock studies. Sulfides in mantle rocks crystallised from potential metasomatic agents including pyroxenites, amphibole veins, and amphibole-apatite xenoliths, have much lower Au concentrations (typically <0.05 ppm). A chromatographic metasomatic model is proposed to explain the variations in sulfide abundance and composition. Peridotitic sulfides show a systematic change in Au content related to their proximity to a metasomatic conduit, as inferred from changes in the (La/Yb)n of clinopyroxene and abundances of volatile-bearing phases. Sulfides in peridotites adjacent to the conduits are more homogenous and have lower Au contents than those distal to the conduit. The homogeneity of these sulfides is interpreted to result from extensive equilibration with the metasomatic agent; Au also has partitioned into the fluid phase, reducing the Au content of these sulfides. The least metasomatised peridotites (“anhydrous” harzburgites with (La/Yb)n <1) have a more diverse sulfide assemblage, generally with higher Au contents. The agents that have affected these samples have been modified by extensive fractionation and wall-rock interaction during percolation through the mantle, becoming enriched in Au. Au is highly mobile in the mantle; it partitions into a fluid phase, and is only deposited during the last stages of fractionation. This has implications for mobilisation of Au from the mantle into the crust. Transient melts and/ or fluids could potentially strip the lithospheric mantle of its Au content as they pass through, and transfer it into the overlying crust. The efficiency of this process is probably more important for the Au-mineralisation process than the Au endowment of the source, because the fractal-scale heterogeneities outlined above imply that Au-rich and Au-poor mantle domains are likely to co-exist in any single location.