Many rifts develop in response to multiphase extension with numerical and physical models suggesting that reactivation of first-phase normal faults and rift-related variations in bulk crustal rheology control the evolution and final geometry of subsequent rifts. However, many natural multiphase rifts are deeply buried and thus poorly exposed in the field and poorly imaged in seismic reflection data, making it difficult to test these models. Here we integrate recent 3D seismic reflection and borehole data across the entire East Shetland Basin, northern North Sea, to constrain the long-term, regional development of this multiphase rift. We document the following key stages of basin development: (i) pre-Triassic to earliest Triassic development of multiple sub-basins controlled by widely distributed, NNW- to NE-trending, east- and west-dipping faults; (ii) Triassic activity on a single major, NE-trending, west-dipping fault located near the basins western margin, and formation of a large half-graben; and (iii) Jurassic development of a large, E-dipping, N- to NE-trending half-graben near the eastern margin of the basin, which was associated with rift narrowing and strain focusing in the Viking Graben. In contrast to previous studies, which argue for two discrete periods of rifting during the Permian–Triassic and Late Jurassic–Early Cretaceous, we find that rifting in the East Shetland Basin was protracted from pre-Triassic to Cretaceous. We find that, during the Jurassic, most pre-Jurassic normal faults were buried and in some cases cross-cut by newly formed faults, with only a few being reactivated. Previously developed faults thus had only a limited control on the evolution and geometry of the later rift. We instead argue that strain migration and rift narrowing was linked to the evolving thermal state of the lithosphere, an interpretation supporting the predictions of lithosphere-scale numerical models. Our study indicates that additional regional studies of natural rifts are required to test and refine the predictions of physical and numerical models, more specifically, our study suggests models not explicitly recognising or including thermal or rheological effects might over emphasise the role of discrete pre-existing rift structures such as normal faults.