We have examined the high quality data of 306 mid-ocean ridge basalt (MORB) glass samples from the East Pacific Rise (EPR), near-EPR seamounts, Pacific Antarctic Ridge (PAR), near-PAR seamounts, Mid-Atlantic Ridge (MAR), and near-MAR seamounts. The data show a correlated variation between Eu/Eu⁎ and Sr/Sr⁎, and both decrease with decreasing MgO, pointing to the effect of plagioclase crystallization. The observation that samples with MgO>9.5 wt.% (before plagioclase on the liquidus) show Eu/Eu⁎>1 and Sr/Sr⁎>1 and that none of the major phases (i.e., olivine, orthopyroxene, clinopyroxene, spinel and garnet) in the sub-ridge mantle melting region can effectively fractionate Eu and Sr from otherwise similarly incompatible elements indicates that the depleted MORB mantle (DMM) possesses excess Sr and Eu, i.e., [Sr/Sr⁎]DMM>1 and [Eu/Eu⁎]DMM>1. Furthermore, the well-established observation that DNb≈DTh, DTa≈DU and DTi≈DSm during MORB mantle melting, yet primitive MORB melts all have [Nb/Th]PMMORB>1, [Ta/U]PMMORB>1 and [Ti/Sm]PMMORB>1 (where PM indicates primitive mantle normalized), also points to the presence of excess Nb, Ta and Ti in the DMM, i.e., [Nb/Th]PMDMM>1, [Ta/U]PMDMM>1 and [Ti/Sm]PMDMM>1. The excesses of Eu, Sr, Nb, Ta and Ti in the DMM complement the well-known deficiencies of these elements in the bulk continental crust (BCC). These new observations, which support the notion that the DMM and BCC are complementary in terms of the overall abundances of incompatible elements, offer new insights into the crust–mantle differentiation. These observations are best explained by partial melting of amphibolite of MORB protolith during continental collision, which produces andesitic melts with a remarkable compositional (major and trace element abundances as well as key elemental ratios) similarity to the BCC, as revealed by andesites in southern Tibet produced during the India–Asia continental collision. An average amphibolite of MORB protolith consists of ~66.4% amphibole, ~29.2% plagioclase and 4.4% ilmenite. In terms of simple modal melting models, the bulk distribution coefficient ratios D2Eu/(Sm+Gd)=1.21, D2Sr/(Pr+Nd)=1.04, DNb/Th=44, DTa/U=57, DTi/Sm=3.39 and DNb/Ta=1.30 readily explains the small but significant negative Eu and Sr anomalies, moderate negative Ti anomaly and huge negative Nb and Ta anomalies as well as the more sub-chondritic Nb/Ta ratio in the syncollisional andesitic melt that is characteristic of and contributes to the continental crust mass. These results support the hypothesis that continental collision zones are primary sites of net continental crust growth, whereas the standard “island arc” model has many more difficulties than certainties. That is, it is the continental collision (vs. "island arc magmatism" or "episodic super mantle avalanche events") that produces and preserves the juvenile crust, and hence maintains net continental growth. The data also allow us to establish the robust composition of depleted and most primitive (or “primary”) MORB melt with 13% MgO. This, together with the estimated positive Eu and Sr anomalies in the DMM, further permits estimation that the DMM may occupy the uppermost ~680 km of the convective mantle following the tradition that the DMM lies in the shallowest mantle. However, the tradition may be in error. The seismic low velocity zone (LVZ) may be compositionally stratified with small melt fractions concentrated towards the interface with the growing lithosphere because of buoyancy. Such small melt fractions, enriched in volatiles and incompatible elements, continue to metasomatize the growing lithosphere before it reaches the full thickness after ~70 Myrs. Hence, the oceanic mantle lithosphere is a huge enriched geochemical reservoir. On the other hand, deep portions of the LVZ, which are thus relatively depleted, become the primary source feeding the ridge because of ridge-suction-driven lateral material supply to form the crust and much of the lithosphere at and in the vicinity of the ridge.