In order to study the origin of the Grande Ronde basalts (GRs) erupted in the climax stage of the Columbia River basalts (CRBs), we carried out high pressure melting experiments on four of the most primitive rock compositions representing the Yakima group of the CRBs. The voluminous GRs (constituting >80 vol% of CRBs) are totally aphyric basaltic andesites. GRs show very narrow and coherent chemical trends both in major and trace elements as well as isotopes. The silica-rich GRs (SiO₂ = 52–56 wt%) can be produced by direct partial melting of a MORB like source material (CRB72-31) at ∼2 GPa or ∼70 km depth. By 30–50% partial melting of the CRB72-31, the entire compositional range of the GRs can be produced in a narrow temperature interval (1300–1350°C) at ∼2 GPa. The aluminous clinopyroxene that appears in the above melting range is consistent as the major controlling phase of the GR trends. The partial melts are very similar to the GRs except for Al₂O₃ and FeO which could be due to the mismatch in the source rock composition. Judging from the variation in REE, involvement of garnet in GR magma genesis can be ruled out. Small amounts of plagioclase (10–30 wt%) may be present in the partial melting residue. Judging from REE patterns and Nd isotopes of the GRs, the source rock should be unfractionated in REE. Based on the melting experiments, a heterogeneous plume model is proposed for the initial stage of the Yellowstone hot spot. Large lithologically distinct blobs of old oceanic crust components were included in the plume head. The GR magmas were produced by partial melting of the oceanic crust components at the bottom of the North American lithosphere. Similar melting processes of basalt/peridotite composite source may be operating in other LIPs (large igneous provinces). The GR type genuine oceanic crust derived melts may be seen where the ambient peridotite remains under subsolidus conditions. Volume and temperature of mantle plumes may have been overestimated, because contributions from the recycled oceanic crust is so large and the current mantle melting models concern only peridotite source.