The Upper Critical Zone of the Bushveld Complex, South Africa, has been divided into so-called cyclic units. Ideally, they should consist of (from the base upward) layers of chromitite, pyroxenite, norite and anorthosite. Some hypotheses for their origin suggest that they result from the addition and mixing of a magnesian magma having orthopyroxene (or olivine) as liquidus mineral. These models predict that the bulk cumulate composition in a cyclic unit formed from such magma ought to contain excess orthopyroxene to plagioclase relative to their cotectic proportions—an hypothesis tested here. All rocks investigated here from the Upper Critical Zone at Impala Platinum Mine are mixtures of cumulus plagioclase and orthopyroxene (except for the chromitites), with very minor intercumulus minerals, mainly clinopyroxene. Hence, the proportion of plagioclase can be determined by measurement of the density of a sample. The density of every piece of continuous bore core from the Upper Group 2 (UG2) and Merensky cyclic units has been determined, and the proportion of plagioclase calculated. For the Bastard cyclic unit, estimates of the proportion of plagioclase have been made based on published modal and chemical analyses. Vertical summation of the proportion of plagioclase indicates that these three cyclic units—UG2, Merensky and Bastard, respectively—contain 64, 74 and over 80% plagioclase by weight. For the Merensky and Bastard cyclic units, such proportions exceed that of the plagioclase-orthopyroxene cotectic, and could not have been produced by processes involving magnesian magma. Various hypotheses for the origin of the excess plagioclase in these cyclic units are discussed. It has been argued that retention of plagioclase in the magma from previous cycles, which is then deposited in an overlying cycle, may have occurred. However, in the case of the Merensky and Bastard cyclic units, the initial isotopic Sr ratios are significantly higher than in their footwall successions, limiting the extent to which this process could have occurred. The hypothesis that two magmas are involved in the formation of each cyclic unit, and that all anorthosite layers in the Critical Zone are derived from a plagioclase-saturated magma, is not considered applicable to most of the cyclic units in view of chemical consanguinity between anorthosites and underlying rock types. The hypothesis proposed here is a significant modification on that model, and envisages the addition at the level of the Merensky Reef of a relatively evolved, basic magma that had plagioclase as the liquidus mineral. It was emplaced as a basal flow underneath the resident magma, from which orthopyroxene continued to crystallize and sink to produce the basal layer of pyroxenite. Accumulation of the combined plagioclase from both liquid layers produced an excess of plagioclase in both the Merensky and Bastard cyclic units. A further conclusion of the study of the UG2 cyclic unit is that a 40 m thick vertical section of homogeneous norite with cotectic proportions accumulated, suggesting no sorting of phases. In contrast, in other sections, systematic vertical changes in proportions of pyroxene and plagioclase are recorded, suggesting periods when mineral sorting was operative.