The isotopes Xe-129, produced from the radioactive decay of extinct I-129, and Xe-136, produced from extinct Pu-244 and extant U-238, have provided important constraints on early mantle outgassing and volatile loss from Earth(1,2). The low ratios of radiogenic to non-radiogenic xenon (Xe-129/Xe-130) in ocean island basalts (OIBs) compared with mid-ocean-ridge basalts (MORBs) have been used as evidence for the existence of a relatively undegassed primitive deep-mantle reservoir(1). However, the low Xe-129/Xe-130 ratios in OIBs have also been attributed to mixing between subducted atmospheric Xe and MORB Xe, which obviates the need for a less degassed deep-mantle reservoir(3,4). Here I present new noble gas (He, Ne, Ar, Xe) measurements from an Icelandic OIB that reveal differences in elemental abundances and Ne-20/Ne-22 ratios between the Iceland mantle plume and the MORB source. These observations show that the lower Xe-129/Xe-130 ratios in OIBs are due to a lower I/Xe ratio in the OIB mantle source and cannot be explained solely by mixing atmospheric Xe with MORB-type Xe. Because I-129 became extinct about 100 million years after the formation of the Solar System, OIB and MORB mantle sources must have differentiated by 4.45 billion years ago and subsequent mixing must have been limited. The Iceland plume source also has a higher proportion of Pu- to U-derived fission Xe, requiring the plume source to be less degassed than MORBs, a conclusion that is independent of noble gas concentrations and the partitioning behaviour of the noble gases with respect to their radiogenic parents. Overall, these results show that Earth's mantle accreted volatiles from at least two separate sources and that neither the Moon-forming impact nor 4.45 billion years of mantle convection has erased the signature of Earth's heterogeneous accretion and early differentiation.