Melt extraction is a process with a length scale that spans many orders of magnitude. Studies of residual migmatites and granulites suggest that melt has migrated from grain boundaries to networks of leucosome-filled structures to steeply inclined cylindrical or tabular granites inferred to have infilled ascent conduits. For example, in anatectic rocks from southern Brittany, France, during decompression-induced biotite-breakdown melting, melt is inferred to have been expressed from foliation-parallel structures analogous to compaction bands to dilation and shear bands, based on location of residual leucosome, and from this network of structures to ascent conduits, preserved as dykes of granite. The leucosome-filled deformation band network is elongated parallel to a sub-horizontal lineation, suggesting that mesoscale melt flow was focused primarily in the plane of the foliation along the lineation to developing dilatant transverse structures. The leucosome network connects with petrographic continuity to granite in dykes; however, the orientation of dykes discordant to fabric anisotropy suggests that their formation was controlled by stress, which indicates that the process is a fracture phenomenon. Blunt fracture tips and zigzag propagation paths indicate that the dykes represent ductile opening-mode fractures; these are postulated to have formed by coalescence of melt pockets. The structures record a transition from accumulation to draining; quantitative volume fluxes are calculated and presented for the generalised extraction process. The anatectic system may have converged to a critical state at some combination of melt fraction and melt distribution that enabled formation of ductile opening-mode fractures, but fractal distribution of inferred mesoscale melt-filled structures has not been demonstrated; this may reflect the inherent anisotropy and/or residual nature of the drained source. Melt extraction has been modelled as a self-organised critical phenomenon, but the mechanism of extraction is not described and the relationship between these models and the spatial and temporal granularity of lower continental crust is not addressed. Self-organised critical phenomena are driven systems involving ‘avalanches’ with a fractal frequency-size distribution; thus, the distribution of melt batch sizes might be expected to be fractal, but this has not yet been demonstrated in nature.