Sulfur isotopes are a powerful geochemical tracer in high-temperature processes, but have rarely been applied to the study of mantle metasomatism. In addition, there are very limited S isotope data on sub-continental lithospheric mantle (SCLM) material. For cratonic regions, these data are restricted to sulfide inclusions in diamonds. To provide new constraints on the S isotope composition of the SCLM and on the source(s) of mantle metasomatic fluids beneath the diamondiferous Kimberley region (South Africa), we investigated the S isotope systematics of five metasomatised mantle xenoliths from the Bultfontein kimberlite. Pentlandite and chalcopyrite in these xenoliths were analysed by in situ secondary-ion mass spectrometry (SIMS), with bulk-rock material measured by gas source isotope ratio mass spectrometry techniques. Based on previous studies, the xenoliths experienced different types of metasomatism to one another at distinct times (∼180 and ∼90-80 Ma). Contained pentlandite grains show variable alteration to heazlewoodite (i.e. Ni sulfide) + magnetite. The in situ S isotope analyses of pentlandite exhibit a relatively restricted range between -5.9 and -1.4‰δ34S (compared to VCDT), with no statistically meaningful differences between samples. Chalcopyrite only occurs in one sample and shows δ34S values between -5.4 and -1.0‰. The bulk-rock Ssulfide isotope analyses vary between -3.4 and +0.8‰δ34S. Importantly, the only sample hosting dominantly fresh sulfides shows a bulk-rock δ34S value consistent with the mean value for the sulfides, whereas the other samples exhibit higher bulk 34S/32S ratios. The differences between bulk-rock and average in situ δ34S values are directly correlated with the degree of sulfide alteration. This evidence indicates that the elevated 34S/32S ratios in the bulk samples are not due to the introduction of heavy S (commonly as sulfates) and are best explained by isotopic fractionation coupled with the removal of light S during serpentinisation, when pentlandite is altered to S-poor mixtures of heazlewoodite and magnetite. Available bulk Ssulfide isotopic data for SCLM peridotite xenoliths are dominated by positive δ34S values, which contrasts with the negative values of the sulfides. These results imply that the mantle S isotope values from bulk peridotite samples are commonly modified by isotopic fractionation during serpentinisation. Therefore, the S isotopic composition of the SCLM may require revision. The limited isotopic variability shown by sulfides in the Bultfontein mantle xenoliths is probably due to intermittent tapping of a mantle source with a relatively restricted S isotope composition. While the asthenospheric mantle (δ34S≤-1.4‰) is a viable candidate, δ34S values as low as -5.9‰ require input from recycled crustal material, possibly represented by the sulfur reservoir with negative δ34S signature that is missing in the >500 Ma sedimentary record and could have been subducted and stored in the Earth's mantle.