The origin of granitic rocks has been the subject of intense debate. It is broadly accepted that granites are generated in the continental crust. However, the composition of the source rocks and the processes controlling melting reactions remain controversial. Fluid-absent melting is a plausible mechanism for granite generation but is not supported by experimental data. Fluid-present melting is favoured by experiments but the presence of a free water phase is very unlikely in the deep crust. Isotopic ratios are also paradoxical because Nd isotopes indicate a crustal (recycled) source but the low 87Sr/86Sr ratios displayed by many granite batholiths require a substantial mantle contribution. Furthermore, mantle–crust hybridization mechanisms are difficult to apply in terms of magma mixing and assimilation according to models of the physical/rheological behaviour of silicic magmas. Here I present a new, source-based, genetic approach for granitic rocks in which all these paradoxes are satisfactorily explained. The model is supported by new constraints from laboratory experiments and observations of field relationships in the Lewisian complex of northern Scotland. The process starts with the invasion of crustal regions by water and K-rich monzodiorite (andesite) magmas derived from hybridized regions of the subduction-modified mantle wedge. These intermediate magmas supply the water and potassium necessary for crustal melting and granite batholith generation.