The genesis of the pegmatitic pyroxenite that often forms the base of the Merensky Unit in the Bushveld Complex is re-examined. Large (>1 cm) orthopyroxene grains contain tricuspidate inclusions of plagioclase, and chains and rings of chromite grains, which are interpreted to have grown by reaction between small, primary orthopyroxene grains and superheated liquid. This superheated liquid may have been an added magma or be due to a pressure reduction as a result of lateral expansion of the chamber. There would then have been a period of non-accumulation of grains, permitting prolonged interaction with the crystal mush at the crystal–liquid interface. Crystal ageing and grain enlargement of original orthopyroxene grains would ensue. Only after the pegmatitic pyroxenite had developed did another layer of chromite and pyroxenite, with normal grain size, accumulate above it. Immiscible sulphide liquids formed with the second pyroxenite, but percolated down as a result of their density contrast, even as far as the footwall anorthosite in some cases. Whole-rock abundances of incompatible trace elements in the pegmatitic pyroxenite are comparable with or lower than those of the overlying pyroxenite, and so there is no evidence for addition and/or trapping of large proportions of interstitial liquid, or of an incompatible-element enriched liquid or fluid in the production of the pegmatitic rock. Because of the coarse-grained nature of the rock, modal analysis, especially for minor minerals, is unreliable. Annealing has destroyed primary textures, such that petrographic studies should not be used in isolation to distinguish cumulus and intercumulus components. Geochemical data suggest that the Merensky pyroxenite (both pegmatitic and non-pegmatitic) typically consists of about 70–80% cumulus orthopyroxene and 10–20% cumulus plagioclase, with a further 10% of intercumulus minerals, and could be considered to be a heteradcumulate.