Description: The Upper Critical Zone (UCZ) of the Bushveld Igneous Complex displays spectacular layering in the form of cyclic units comprising a basal chromitite layer overlain by a sequence of silicate cumulates in the order, from bottom to top, pyroxenite–norite–anorthosite. Electron microprobe and laser ablation inductively coupled plasma mass spectrometry analyses of chromite and silicate minerals in layers between the UG2 chromitite and the Merensky Reef reveal variations in major and trace element compositions that defy explanation with existing models of cumulate mineral–melt evolution. The anomalous features are best developed at sharp contacts of chromitite with adjacent anorthosite and pyroxenite cumulates. Here, chromite compositions change abruptly from high and constant Mg/(Mg + Fe 2+ ) and Fe 2+ /Fe 3+ ratios in chromitite layers to variable and generally lower values in chromite disseminated in silicate layers. Furthermore, the composition of disseminated chromites varies depending on the host silicate assemblage; for example, in Ti, V and Zn contents. Importantly, the abrupt change in chromite composition across the chromitite–silicate layer contacts is independent of the thickness of the chromitite layer and the estimated mass proportions of chromite to intercumulus liquid. Chemical variations in plagioclase are also abrupt and some are hard to reconcile with conventional models of re-equilibration with intercumulus liquid. Among those features is the decoupling of alkalis from other incompatible lithophile elements. In comparison with cumulus plagioclase, intercumulus poikilitic plagioclase in chromitite layers is enriched in rare earth elements but strongly depleted in equally incompatible Li, K and Rb. Strong alkali depletion is also observed in intercumulus pyroxene from ultramafic cumulates and chromitite layers. To explain these features, we propose a new model of post-cumulus recrystallization, which intensifies the modal layering in the crystal–liquid mush, producing the observed sequence of nearly monomineralic layers of chromitite, pyroxenite and anorthosite that define the cyclic units. The crucial element of this model is the establishment of redox potential gradients at contacts between chromite-rich cumulates and adjacent silicate layers owing to peritectic reactions between the crystals and intercumulus melt. Because basaltic melts are ionic electrolytes with Na + as the main charge carrier, the redox potential gradient induces electrochemical migration of Na + and other alkali ions. Selective mobility of alkalis can explain the enigmatic features of plagioclase composition in the cyclic units. Sodium migration is expected to cause remelting of previously formed cumulates and major changes in modal mineral proportions, which may eventually result in the formation of sharply divided monomineralic layers. The observed variations in ferric/ferrous iron ratios in chromite from the cyclic units and Fe distribution in plagioclase imply a redox gradient of the order of 0·9 log-units f O 2 , equivalent to a potential gradient of 60 mV. Preliminary estimates suggest that the resulting electrochemical flux of Na + ions is sufficient to mobilize about one-third of the total Na content of a 1 m thick mush layer within 10 years. The proposed electrochemical effect of post-cumulus crystallization is enhanced by the presence of cumulus chromite but, in principle, it can operate in any type of cumulates in which ferrous and ferric iron species are distributed unequally between crystalline and liquid phases.