New chemical data show that the two main granite types (G1 and G2) cannot be discriminated, but that microgranite sheets/dykes (G3) are significantly different and more evolved, largely as a result of biotite, accessory mineral, and plagioclase fractionation. The Lundy granite is similar to other Tertiary granites of Scotland and Ireland, in age, setting, possible high-temperature mineralogy, relationship to basic magmatism, and REE patterns. These features and a highly evolved chemistry suggest derivation from an unexposed more ‘primitive’ granite that, in turn, had a basaltic parentage. However, similarities with the nearby S-type Hercynian granites, such as high aluminium saturation index (and normative corundum), high trace alkali, Nb, and F contents, low Zr, and high initial Sr ratio suggest a significant crustal component. The problem is resolved by proposing either mixing of silicic magma derived by strong fractionation of basaltic magma with anatectic magma from a pelitic/semi-pelitic crustal source, or fractionation of basaltic magma heavily contaminated by assimilated crustal material. Both origins would yield the high REE contents and fiat REE patterns of a ‘primitive’ granite magma. Fractionation, perhaps of hornblende initially, and later, of biotite and accessory minerals together with feldspars, would produce the small volume of highly fractionated Lundy granite.