Abstract Zabargad Island in the Red Sea exposes ultramafic bodies in contact with partially retrograded granulite-facies lower crustal rocks; the contact could represent the sub-Red Sea Moho prior to the opening of the Red Sea. The ultramafic rocks are largely protogranular spinel lherzolites that grade locally to porphyroclastic to mylonitic amphibole-peridotites. Some of the peridotites are infiltrated by high Mg-gabbros (troctolites). The peridotites and the lower crustal rocks are cut by diabase dikes. The Zabargad peridotites are isotopically heterogeneous on a scale of meters or less. Sr and Nd isotopic ratios of clinopyroxenes (cpx) from spinel lherzolites lie in the depleted portion of the 143Nd 144Nd − 87Sr 86Sr “mantle array” whereas cpx and amphiboles from amphibole-peridotites fall off the array towards the low 143Nd 144Nd and sol 87Sr 86Sr field that may characterize portions of the sub-continental mantle. Cpx from troctolites plot in the middle of the 143Nd 144Nd - 87Sr 86Sr mantle array and have higher 87Sr 86Sr ratios and more depleted LREE patterns than the intruded spinel-lherzolites and amphibole peridotites. All whole-rock lherzolites and peridotites and most diabase dikes from Zabargad, as well as a variety of rocks from other parts of the Red Sea area (MORB glasses, a diabase dike from The Brothers Island, basanites and clinopyroxenes from associated nodules from the Ataq diatreme in Yemen) define a linear array on a Sm-Nd isochron diagram with a best-fit “age” of ca. 675 Ma and an initial ϵ Nd of +6.7. This array is probably a mixing line between depleted MORB-type mantle and a low 143Nd 144Nd − 87Sr 86Sr metasomatic component. The “age” defined by the linear array may date either an episode of Pan-African mantle homogenization or, in our preferred model, the Pan-African segregation of a previously homogeneous mantle into two or more reservoirs of identical 143Nd 144Nd , but different Sm Nd ratios. The subsequent partial remixing of these isotopic reservoirs during metasomatism associated with the opening of the Red Sea did not perturbed the slope of the array. The LREE-enriched reservoir may have been the Pan-African lower crust since there is field and textural evidence that the metasomatic fluids were derived from, or equilibrated with, the adjacent crustal complex. Alternatively, the LREE-enriched component may have been a mantle reservoir similar to that which generated basalts and nodules from the nearby Ataq diatreme. The troctolites are isotopically distinct from the host peridotites and retain igneous textures suggesting they intruded the peridotites during the opening of the Red Sea. Their isotopic patterns are similar to those of recent volcanics from Afar, portions of the Ataq diatreme and some Zabargad dikes indicating derivation from a similar mantle source. Their very LREE-depleted patterns suggest they are either re-melted fractions of previously very depleted spinel lherzolites or early-formed mineral aggregates that became enmeshed in the peridotites as fractionating melt ascended to higher levels. Our results indicate that at least two major processes (metasomatism and intrusion) have modified the isotopic-pat-terns of the spinel lherzolites of Zabargad. Had the resultant chemically and isotopically heterogeneous bodies stayed in the mantle, instead of being uplifted to their present positions, they would have proven to be enigmatic sources for future basalts and xenoliths.