A Deep Hot Zone develops when numerous mafic sills are repeatedly injected at Moho depth or are scattered in the lower crust. The melt generation is numerically modelled for mafic sill emplacement geometries by overaccretion, underaccretion or random emplacement, and for intrusion rates of 2, 5 and 10 mm/yr. After an incubation period, melts are generated by incomplete crystallisation of the mafic magma and by partial melting of the crust. The first melts generated are residual from the mafic magmas that have low solidi due to concentration of H 2 O in the residual liquids. Once the solidus of the crust is reached, the ratio of crustal partial melt to residual melt increases to a maximum. If wet mafic magma, typical of arc environments, is injected in an amphibolitic crust, the residual melt is dominant over the partial melt, which implies that the generation of I-type granites is dominated by the crystallisation of mafic magma originated from the mantle and not by the partial melting of earlier underplated material. High ratios of crustal partial melt over residual melt are reached when sills are scattered in a metasedimentary crust, allowing the generation of S-type granites. The partial melting of a refractory granulitic crust intruded by dry, high-T mafic magma is limited and subordinate to the production of larger amount of residual melt in the mafic sills. Thus the generation of A-type granites by partial melting of a refractory crust would require a mechanism of selective extraction of the A-type melt.