Uranium series disequilibria in ocean island basalts (OIB) provide evidence for the presence of garnet in their source region. It has been suggested that enriched OIB signatures derive from mantle lithologies other than peridotite, such as eclogite or pyroxenite, and, in particular, that silica-poor garnet pyroxenite is the source lithology for alkali basalts. To test the ability of such a source to produce the U–Th disequilibria observed in alkali OIB, we determined experimentally clinopyroxene-melt and garnet-melt partition coefficients for a suite of trace elements, including U and Th, at 2.5 GPa and 1420–1450 °C. The starting composition for the experiments was a 21% partial melt of a silica-poor garnet pyroxenite. Experimentally determined clinopyroxene-melt partition coefficients range from 0.0083 ± 0.0006 to 0.020 ± 0.002 for Th and from 0.0094 ± 0.0006 to 0.024 ±0.002 for U, and garnet-melt partition coefficients are 0.0032 ± 0.0004 for Th and 0.013 ± 0.002 for U. Comparison of our experimental results with partition coefficients from previous experimental studies shows that the relative compatibilities of U and Th in both garnet and clinopyroxene are different for different mineral compositions, leading to varying degrees of U/Th fractionation with changing lithology. For a given melting rate and extent of partial melting, mafic lithologies tend to produce larger 230Th excesses than peridotite. However, this effect is minimized by the greater overall extents of melting experienced by eclogites and pyroxenites relative to peridotite. Results from chromatographic, batch, and fractional melting calculations with binary mixing between partial melts of pyroxenite and peridotite, carried out using our new partitioning data for the pyroxenite component and taking into account variable productivities and different solidus depths for the two lithologies, suggest that OIB are not the product of progressive melting of a source containing a fixed quantity of garnet pyroxenite. Melting a peridotite with enriched signatures, and mixing those melts with melts of a depleted, “normal” peridotite, is an alternative explanation for the trends seen in Hawaiian, Azores and Samoan lavas.