Description: Iron isotopes in ocean floor basalts (OFB) away from convergent margins comprising mid-ocean-ridge and ocean island lavas show significant variation of 〉0.4‰ (expressed in the delta notation δ57Fe relative to IRMM-014), but processes responsible for this variation remain elusive. Bond-valence theory predicts that valence states (Fe3+ vs. Fe2+) control Fe isotopes during partial melting and crystal fractionation along the liquid line of descent and thus contribute substantially to this variation. Memory of past melt extraction or metasomatic re-enrichment in the source of OFB may further add to the observed variability, but systematic investigations to elucidate the respective contributions of these effects have been lacking. Submarine ridges and rifts in the Lau back-arc basin offer a unique opportunity to compare Fe isotopes in OFB from different melting regimes and variably depleted mantle sources. New Fe isotope data is presented for submarine lavas from the Rochambeau Ridges (RR) and the Northwest Lau Spreading Centre (NWLSC), and is compared with published data from the Central Lau Spreading Centre (CLSC). In line with first principle calculations and observations from a range of natural systems, crystal fractionation is identified as the dominant, controlling process for elevating δ57Fe in the lavas with olivine tentatively identified as the key driver. To compensate for the effect of crystal fractionation, olivine is mathematically added towards calculated primitive melt compositions (δ57Feprim). For this, we used a constant Ol-melt isotope fractionation factor based on published equilibrium partition functions adapted to decreasing temperature in a cooling melt. The degree of calculated Fe isotope fractionation through olivine crystal fractionation (monitored as Δ57Fe = δ57Femeasured − δ57Feprim) is positively correlated with increasing S and decreasing Ni content in the cooling lavas, fortifying the validity of the approach. Primitive lavas from individual Lau spreading centres and ridges vary to 0.1‰ in δ57Feprim, similar to primitive open-ocean MORB. However, the entire spread in Fe isotope variability in the primitive melts remains at 0.3‰, which we propose to be the extent of isotope heterogeneity in Earth’s upper mantle, with few extreme exceptions. The largest variability in δ57Feprim is observed for RR intra-plate lavas, which have been associated with the Samoan mantle plume and melting in an edge-driven convection scenario. Low, mid-ocean ridge-like 87Sr/86Sr in RR lavas excludes significant influence of isotopically heavy Samoan EM2-type components. However, co-variations with rare earth element pattern in some RR intra-plate lavas indicate garnet plays a role in elevating δ57Feprim during deeper melting. Excluding these deep-seated melts uncovers systematically decreasing δ57Feprim coupled to the degree of mantle source depletion, as recorded in Lu/Hf and Sm/Nd, in the back-arc basin basalts. This, however, holds only true for a comparison between sources of individual ridges, whereas no co-variation is observed within ridge segment data. This suggests that a process other than source depletion and crystal fractionation further adds to Fe isotope variability in the order of 0.1‰ on scales of individual ridge segments. This either marks the degree of Fe isotope variability below ridge segments, or is caused by secondary processes, such as melt-wallrock interaction or RTX (recharge and crystal fractionation) magma chambers.