Since the first reconstruction of the supercontinent Pangaea, key advances in plate tectonic reconstructions have been made [1][2][3][4][5][6] . Although the movement of tectonic plates since the start of the mid-Cretaceous period (∼100 million years (Myr) ago) is relatively well understood 1,2 , the longitudinal position of plates before this period is not constrained at all. Here, we use a global mantle tomography model 7 to estimate the longitude of past oceanic subduction zones. We identify 28 remnants of oceanic plates that were subducted into the lower mantle and link these to the mountain building zones from which they are likely to have originated. Assuming that these remnants sank vertically through the mantle, we reconstruct the longitude at which they were subducted. Our estimates for the location of the subduction zones are offset by up to 18• compared with plate tectonic reconstructions for the corresponding period. We did not detect oceanic plate remnants from the Carboniferous period (∼300-360 Myr ago), or before, suggesting that the tomographic visibility of subduction is limited to the past 300 Myr.Since the first qualitative plate reconstruction of the supercontinent Pangaea was determined by fitting palaeoclimatic belts and modern continental margins, key advances in plate reconstructions have been made with the development and use of palaeomagnetic apparent polar wander paths, ocean floor magnetic anomalies and hotspot reference frames 1,2 , leading to global plate tectonic reconstructions 3,5,6 . Absolute plate motion models have often been based on assumed hotspot fixity and are well constrained only up to the Cretaceous period owing to the lack of any preserved older oceanic hotspot tracks 1,2 . These, and other models, offer no control on absolute palaeolongitude before the Cretaceous.Seismic tomography studies of the mantle have allowed for increasingly detailed correlations between deep mantle structure, mostly focused on presumed remnants of subducted plates and plate tectonic evolution [8][9][10][11][12][13][14][15][16] . This, however, has not led to strong constraints on absolute plate motion. Recently, correlations between deep, presumably hot and dense mantle heterogeneities at the core-mantle boundary and large igneous provinces were obtained from a plate reconstruction 3,4 , leading to possible predictions of absolute palaeolongitude for the entire Phanerozoic eon 17 . This reconstruction model, however, assumes zero longitude motion for Africa before the Cretaceous.Here, independently of any reconstruction model, we carry out a global interpretation of positive seismic anomalies in the lower mantle based on the assumptions that these reflect relatively