Petrological studies of Siberian Traps and associated alkaline rocks reveal high temperatures (1600-1650 o C) 14,17 in their mantle sources. Olivine compositions in samples from lower units of the Norilsk lava section provide evidence that the mantle source of the Siberian Traps was unusually rich in ancient recycled oceanic crust 14 in agreement with earlier predictions 10 . For the main volcanic phase, however, such data were unavailable. Here we report 2500 new olivine analyses and host-rock compositions for 45 basalts covering the main stages of 3 tholeiitic magmatism in three key localities: the Norilsk area, the Putorana plateau and the Maymecha-Kotuy province (Fig. 1). Almost all olivine compositions possess significantly higher NiO and FeO/MnO than expected for olivine in peridotite-derived magmas (Fig.1b,c supplementary Fig. S1), suggesting a contribution of melts from pyroxenitic sources 18 . Alternative explanations of these observations seem less plausible (see Methods for discussion). Our interpretation of the olivine compositions implies that the source of the Siberian Traps contained 10-20 wt.% recycled oceanic crust (Methods). More specifically, all lavas erupted during the first stage of magmatic activity (Gudchikhinskaya and earlier suits of the Norilsk area) are depleted in heavy rare earth elements (HREE) 19,20 indicating residual garnet and derivation within or below the base of thick lithosphere (>130 km) 14 . The source of Gudchikhinskaya lavas was likely almost entirely pyroxenitic 14 (Fig. 1b,c,d).Younger magmas are not depleted in HREE indicating formation at shallow depths and dramatic thinning of the lithosphere. Our calculation suggests that these magmas had a near-constant proportion of pyroxenite-derived melt of about 50% (Fig. 1d, table S1 in Supplementary Information) and were strongly contaminated by the continental crust 20 . Because the main Norilsk section spans less than 1 m.y. 1 , it is likely that the lithosphere was thinned in only a few hundred thousand years.High mantle temperatures over a vast area (Fig.1a) are consistent with the head of a hot mantle plume 6,9,17 . Based on the petrological constraints we develop a thermomechanical model of the interaction of the plume and lithosphere (see Methods). We assume that the plume arrived below the lithosphere at about 253 Ma (model time 0), perhaps near the northern border of the Siberian Shield, where the hottest melts (meimechites) erupted 17 . We further assume that the plume head was hot (Tp=1600°C, 250°C excess temperature) and contained a high content (15 wt.%.) of recycled oceanic crust. In our two-dimensional model, we approximate the plume head by a half-circle of 400 km radius located below cratonic lithosphere of variable thickness corresponding to the margin of the The arrival of a large and hot mantle plume head at the base of the lithosphere has been predicted 6,21 to cause about 0.8-1 km of broad surface uplift per 100°C of plume excess temperature. For a purely thermal plume with an excess temperature of 25...