On a global scale, peridotitic garnet inclusions in diamonds from the subcratonic lithosphere indicate an evolution from strongly sinusoidal REEN, typical for harzburgitic garnets, to mildly sinusoidal or "normal" patterns (positive slope from LREEN to MREEN, fairly flat MREEN-HREEN), typical for lherzolitic garnets. Using the Cr-number of garnet as a proxy for the bulk rock major element composition it becomes apparent that strong LREE enrichment in garnet is restricted to highly depleted lithologies, whereas flat or positive LREE-MREE slopes are limited to less depleted rocks. For lherzolitic garnet inclusions, there is a positive relation between equilibration temperature, enrichment in MREE, HREE and other HFSE (Ti, Zr, Y), and decreasing depletion in major elements. For harzburgitic garnets, relations are not linear, but it appears that lherzolite style enrichment in MREE-HREE only occurs at temperatures above 1150-1200 °C, whereas strong enrichment in Sr is absent at these high temperatures. These observations suggest a transition from melt metasomatism (typical for the lherzolitic sources) characterized by fairly unfractionated trace and major element compositions to metasomatism by CHO fluids carrying primarily incompatible trace elements. Melt and fluid metasomatism are viewed as a compositional continuum, with residual CHO fluids resulting from primary silicate or carbonate melts in the course of fractional crystallization and equilibration with lithospheric host rocks. Eclogitic garnet inclusions show "normal" REEN patterns, with LREE at about 1× and HREE at about 30× chondritic abundance. Clinopyroxenes approximately mirror the garnet patterns, being enriched in LREE and having chondritic HREE abundances. Positive and negative Eu anomalies are observed for both garnet and clinopyroxene inclusions. Such anomalies are strong evidence for crustal precursors for the eclogitic diamond sources. The trace element composition of an "average eclogitic diamond source" based on garnet and clinopyroxene inclusions is consistent with derivation from former oceanic crust that lost about 10% of a partial melt in the garnet stability field and that subsequently experienced only minor reenrichment in the most incompatible trace elements. Based on individual diamonds, this simplistic picture becomes more complex, with evidence for both strong enrichment and depletion in LREE. Trace element data for sublithospheric inclusions in diamonds are less abundant. REE in majoritic garnets indicate source compositions that range from being similar to lithospheric eclogitic sources to strongly LREE enriched. Lower mantle sources, assessed based on CaSi-perovskite as the principal host for REE, are not primitive in composition but show moderate to strong LREE enrichment. The bulk rock LREEN-HREEN slope cannot be determined from CaSi-perovskites alone, as garnet may be present in these shallow lower mantle sources and then would act as an important host for HREE. Positive and negative Eu anomalies are widespread in CaSi-perovskites and negative anomalies have also been observed for a majoritic garnet and a coexisting clinopyroxene inclusion. This suggests that sublithospheric diamond sources may be linked to old oceanic slabs, possibly because only former crustal rocks can provide the redox gradients necessary for diamond precipitation in an otherwise reduced sublithospheric mantle.