The Tennis Ball Marker (TBM) is a distinctive lithology that is particularly well developed near the base of the Rustenburg Layered Suite (RLS) on the farm Middelkraal 221 JS, approximately 20 km south of the town of Roossenekal in the Eastern Limb of the Bushveld Complex. The name refers to a texture in which approximately tennis ball-sized spheroidal aggregates of feldspathic orthopyroxenite or melanorite occur within a lighter-coloured gabbronorite host rock. We have identified two well-defined layers in which pyroxenitic spheroids are densely packed, with spheroids more sparsely distributed elsewhere in the host gabbronorite. The TBM at Middelkraal has previously been described as a contact phenomenon where the Main Zone (MZ) has been contaminated by footwall lithologies that include basaltic lavas of the Dullstroom Formation. Our geochemical data, in tandem with new geological mapping, suggest that the TBM is in fact hosted by the Marginal Sill Phase (MSP), and not by the MZ. The MSP is a regional feature that separates the overlying units of the RLS from its floor rocks throughout this region. Sills that make up the MSP were injected on a regional plane of weakness in the primary stratigraphy prior to the intrusion of the remainder of the RLS. The MSP does not represent the chilled carapace of a magma chamber, as implied in previous studies, and there is no evidence of a genetic relationship with the MZ. Our geochemical data indicate, furthermore, that the gabbronorite constituting the dominant lithology of the MSP that hosts the TBM at Middelkraal formed from a sequence of several magma influxes. The TBM is ascribed to one or more subsequent intrusions of pyroxenitic magma into the still hot gabbronoritic host sequence. There are no comparable lithologies reported from other layered intrusions, and the closest analogy to the TBM is the Boulder Bed in the Upper Critical Zone (UCZ) of the Western Bushveld Complex, which has received considerably more attention in the literature than has the TBM. The Boulder Bed has been ascribed variously to in-situ agglomeration of clusters of orthopyroxene resulting from liquid immiscibility, in situ breakup of a pre-existing pyroxenite layer, possibly due to seismic events, or the disaggregation of a late-stage pyroxenitic sill intruding into the magma chamber. We subscribe to the latter mechanism for both the TBM and the Boulder Bed, drawing on similarities with mafic magmatic enclaves, where mafic sills have been shown to have disaggregated on intrusion into earlier bodies of felsic to intermediate igneous rock, or peperites, where magma intruded fluidised sediments.