Transitional fossils informing the origin of turtles are among the most sought-after discoveries in palaeontology. Despite strong genomic evidence indicating that turtles evolved from within the diapsid radiation (which includes all other living reptiles6, 7), evidence of the inferred transformation between an ancestral turtle with an open, diapsid skull to the closed, anapsid condition of modern turtles remains elusive. Here we use high-resolution computed tomography and a novel character/taxon matrix to study the skull of Eunotosaurus africanus, a 260-million-year-old fossil reptile from the Karoo Basin of South Africa, whose distinctive postcranial skeleton shares many unique features with the shelled body plan of turtles2, 3, 4. Scepticism regarding the status of Eunotosaurus as the earliest stem turtle arises from the possibility that these shell-related features are the products of evolutionary convergence. Our phylogenetic analyses indicate strong cranial support for Eunotosaurus as a critical transitional form in turtle evolution, thus fortifying a 40-million-year extension to the turtle stem and moving the ecological context of its origin back onto land8, 9. Furthermore, we find unexpected evidence that Eunotosaurus is a diapsid reptile in the process of becoming secondarily anapsid. This is important because categorizing the skull based on the number of openings in the complex of dermal bone covering the adductor chamber has long held sway in amniote systematics10, and still represents a common organizational scheme for teaching the evolutionary history of the group. These discoveries allow us to articulate a detailed and testable hypothesis of fenestral closure along the turtle stem. Our results suggest that Eunotosaurus represents a crucially important link in a chain that will eventually lead to consilience in reptile systematics, paving the way for synthetic studies of amniote evolution and development.