There is uncertainty about whether the abundant tholeiitic lavas on Hawaii are the product of melt from peridotite or pyroxenite/eclogite rocks. Using a parameterization of melting experiments on peridotite with glass analyses from the Hawaii Scientific Deep Project 2 on Mauna Kea volcano, I show here that a small population of the core samples had fractionated from a peridotite-source primary magma. Most lavas, however, differentiated from magmas that were too deficient in CaO and enriched in NiO (ref. 2) to have formed from a peridotite source. For these, experiments indicate that they were produced by the melting of garnet pyroxenite, a lithology that had formed in a second stage by reaction of peridotite with partial melts of subducted oceanic crust. Samples in the Hawaiian core are therefore consistent with previous suggestions that pyroxenite occurs in a host peridotite, and both contribute to melt production. Primary magma compositions vary down the drill core, and these reveal evidence for temperature variations within the underlying mantle plume. Mauna Kea magmatism is represented in other Hawaiian volcanoes, and provides a key for a general understanding of melt production in lithologically heterogeneous mantle.