Using a new high-resolution dataset, this study presents evidence for short length scale 18O/16O heterogeneity in the mantle source region of young (age ≲12 ka bp) Icelandic basalts. The dataset comprises secondary ion mass spectrometry determinations of 18O/16O in single compositional zones of plagioclase crystals from the primitive Borgarhraun flow in northern Iceland, along with trace and major element data from the same zones. The presence of mantle under Iceland with δ18O below typical mid-ocean ridge basalt (MORB) values of ∼5·5 ± 0·3‰ (VSMOW) has previously been disputed, because variability in δ18O in many Icelandic basalts is also known to be caused by the interaction of basaltic melts with crustal lithologies that have been altered by low-δ18O meteoric water. Primitive basalt flows, such as Borgarhraun, and their macrocrysts are the most likely candidates to retain a mantle δ18O signature. However, the role of crustal processes in generating the low δ18O in olivine crystals from these flows has not unequivocally been ruled out. By making intra-crystal analyses in Borgarhraun plagioclase it has been possible in this study to obtain a detailed record of the chemical and isotopic compositions of the melts that crystallized the plagioclase zones. The variability observed in trace element compositions of the early crystallized anorthitic plagioclase zones (80·9–89·4 mol % anorthite) is firstly shown to arise from melt compositional variability, and equilibrium melt concentrations of Sr, La and Y are then calculated from the crystal concentrations of these elements using carefully selected partition coefficients. The ranges of incompatible trace element ratios (La/Y, Sr/Y) in these equilibrium melts reflect a range of compositions of fractional mantle melts, a result that is in agreement with previous proposals for the cause of variability in trace element indices of Borgarhraun olivine-hosted melt inclusions and clinopyroxene compositional zones. Correlations observed between La/Y and Sr/Y in the melts in equilibrium with the Borgarhraun plagioclase zones and the δ18O of these zones therefore support the hypothesis that the mantle under Iceland is heterogeneous in 18O/16O. Such correlations have not previously been observed in intra-crystal data from Iceland, and provide strong evidence that mantle material with abnormally low δ18O may exist in the form of readily fusible heterogeneities alongside ambient mantle with MORB-like δ18O (≈+5·5‰) on a length scale of <100 km. The lowest δ18O of plagioclase that is attributed to a mantle origin in this study is 4·5 ± 0·4‰, equating to a melt equivalent value of 4·3 ± 0·5‰ or an olivine equivalent value of 3·8 ± 0·5‰.