Very large crystals of magnetite, up to 2 cm, are found in pure magnetitite layers in the upper zone of the Bushveld Complex. Detailed electron microprobe analysis of one of these indicates a vertical compositional zoning of Cr (a highly compatible element in magnetite) quite different from the concentric zonation often found, for example, in feldspars in intrusive rocks. It is shown that these crystals could not have grown to their present size in suspension in the magma chamber. Annealing of many small crystals into a single grain could occur either by the Ostwald ripening process at temperatures close to the liquidus or by subsolidus recrystallization. Alternatively, these grains could have formed directly from the magma by upward growth from the floor of the magma chamber, by a process analogous to crescumulate or heteradcumulate growth but in the absence of other crystallizing phases. All of these models require the observed Cr gradient to be produced during the initial crystallization event and not as a secondary effect.