The incompletely known sauropod Tornieria africana from the Upper Jurassic (Tithonian) of Tendaguru, Tanzania, has for over 80 years been regarded to represent a Gondwanan species of the North American genus Barosaurus Marsh (Morrison Formation: Kimmeridgian-Tithonian), but this identification has recently been questioned. The holotype and referred specimens are redescribed here, and the characters present are reevaluated in light of current knowledge of sauropod phylogeny. Synapomorphies of the skull (prefrontal with triangular posterior process) and anterior caudal vertebrae (procoelous centra, presence of diapophyseal laminae, and presence of a pleurocoel) indicate that the Tendaguru material represents a member of the Diplodocinae (Diplodocoidea, Diplodocidae) and is therefore very closely related to Barosaurus and Diplodocus. It differs from all other diplodocine genera in several characters, such as robust anterior caudals with pleurocoels located in the upper third of the centrum and ventral excavations, and stout hind limb proportions similar to Apatosaurus (tibia:femur length ratio less than 0.64). In a phylogenetic analysis, the African form consistently emerges as the sister taxon to Barosaurus+Diplodocus. Therefore, previous suggestions that Tornieria africana is the available name for this taxon are supported by this analysis. The existence of this form in Gondwana contradicts the idea of Laurentian endemism of the diplodocid clade, and is best explained by a vicariance model of diplodocoid paleobiogeography. This implies an extensive ghost lineage of this group, extending back at least as far as the upper Middle Jurassic.