Considerable progress has been made in the understanding of craton formation and evolution in southern Africa in recent years, thanks to newly available geophysical and geochemical data. Recent tomographic studies and Re-Os analyses of on- and off-craton mantle xenoliths have clearly shown that the Kaapvaal craton is underlain by a, cold, depleted lithosphere at least 200 km thick that was stabilized in the Archaean. Several important questions regarding the consolidation and modification of the lithosphere, however, remain unanswered or highly debated. Firstly, the sub-cratonic lithospheric mantle (SCLM) is distinct from oceanic and most continental mantle in that it is more depleted in magmaphile elements (Fe, Al, Ca, HREE), but at the same time strongly enriched in incompatible trace elements like LILE and LREE. The Kaapvaal SCLM also has a distinctly higher Si/Mg ratio and therefore higher modal orthopyroxene content (at a given Mg/Fe) than oceanic and most continental upper mantle. Resolution of this issue is critical to understanding the mechanism of lithosphere formation. We carried out a detailed petrological and geochemical study on low-temperature peridotite xenoliths from Kimberley (located at the craton core) and northern Lesotho (at the south-eastern craton edge) in order to obtain information about the processes that led to the depletion and re-enrichment of the Kaapvaal SCLM. Samples have been characterized for Re-Os isotope systematics, major and trace element concentrations in whole rocks and minerals, and garnet and clinopyroxene Lu-Hf, Sm-Nd and Rb-Sr isotopes. The combined results require a multistage history of the Kaapvaal mantle. They are most consistent with a model that involves introduction of orthopyroxene and trace element enrichment of the garnets by aqueous fluids, possibly derived from a subducting oceanic slab. This is consistent with the observed enrichment in Si (orthopyroxene) and the Re-Os systematics. Infiltration of hydrous fluids would also increase the degree of mantle melting and could therefore explain the strong major and HREE depletion of the SCLM. The Nd-Hf isotope characteristics of the garnets require the trace element enrichment to be ancient. We therefore suggest that melting and metasomatism of the Kaapvaal SCLM took place in subduction zone settings, probably during amalgamation of smaller pre-existing terranes in the Late Archaean (~2.9 Ga). Trace element and Nd and Hf isotope disequilibrium between garnet and clinopyroxene is preserved in many samples and indicates that garnet and clinopyroxene are not cogenetic. Calculated equilibrium liquids and Hf-Nd isotopic compositions for clinopyroxene suggest that most diopside in the xenoliths studied crystallized from an infiltrating kimberlite- like melt, either during group II kimberlite magmatism at ~200-110 Ma ago (Kimberley) or shortly prior to eruption of the host kimberlite around 90 Ma (N. Lesotho).