Major, minor, and rare-earth element analyses of fine-grained marginal material from 249 dykes of the Matachewan giant radiating swarm show that the range in compositions cannot be produced by any simple fractionation model. Instead a two-stage process is required, whereby melts were modified by both fractionation and assimilation of lower crustal material, before rising and ponding at shallower crustal levels (15–20 km maximum), where a combination of fractional crystallization, plagioclase phenocryst formation, and periodic melt replenishment took place. These shallow crustal chambers acted as feeders for the numerous dykes. One hypothesis compatible with these observations involves an extensive rising plume that would provide heterogeneous melts over a large area accompanied by stresses that could produce a radial pattern of dykes. In a search for indications of lateral magma flow, a comparison of longitudinal variations in composition along four sample traverses across the western subswarm showed no truly consistent trends. However, Hf/Zr values are highest in the traverse nearest the focal region and lowest in the traverse 250 km away, suggesting a change in magma composition with distance from the focus of the swarm, the assumed plume centre. Given the presently available data, the variations cannot be attributed to progressive fractionation during lateral magma flow in the crust nor to a northward increase in amounts of contamination by known lower crustal assimilants, all of which have Hf/Zr ratios higher than the distal values recorded by the dykes. Instead the variations are thought to reflect a possible compositional zonation of the plume head. The distortion of the western subswarm from a simple radiating pattern may have been facilitated by a low viscosity lower crust, softened by Matachewan igneous activity, and is consistent with deformation by the Blezardian orogen at about 2.2–2.4 Ga.