Semaphorins and their receptor plexins constitute a pleiotropic cell-signalling system that is used in a wide variety of biological processes, and both protein families have been implicated in numerous human diseases. The binding of soluble or membrane-anchored semaphorins to the membrane-distal region of the plexin ectodomain activates plexin's intrinsic GTPase-activating protein (GAP) at the cytoplasmic region, ultimately modulating cellular adhesion behaviour. However, the structural mechanism underlying the receptor activation remains largely unknown. Here we report the crystal structures of the semaphorin 6A (Sema6A) receptor-binding fragment and the plexin A2 (PlxnA2) ligand-binding fragment in both their pre-signalling (that is, before binding) and signalling (after complex formation) states. Before binding, the Sema6A ectodomain was in the expected 'face-to-face' homodimer arrangement, similar to that adopted by Sema3A and Sema4D, whereas PlxnA2 was in an unexpected 'head-on' homodimer arrangement. In contrast, the structure of the Sema6A-PlxnA2 signalling complex revealed a 2:2 heterotetramer in which the two PlxnA2 monomers dissociated from one another and docked onto the top face of the Sema6A homodimer using the same interface as the head-on homodimer, indicating that plexins undergo 'partner exchange'. Cell-based activity measurements using mutant ligands/receptors confirmed that the Sema6A face-to-face dimer arrangement is physiologically relevant and is maintained throughout signalling events. Thus, homodimer-to-heterodimer transitions of cell-surface plexin that result in a specific orientation of its molecular axis relative to the membrane may constitute the structural mechanism by which the ligand-binding 'signal' is transmitted to the cytoplasmic region, inducing GAP domain rearrangements and activation.