Amniotes include mammals, reptiles and birds, representing 75% of extant vertebrate species on land. They originated around 318 million years ago (early late Carboniferous)and their early fossil record is central to understanding the expansion of vertebrates in terrestrial ecosystems. We present a new phylogenetic hypothesis that challenges the widely-accepted consensus about early amniote evolution, based on parsimony analysis and Bayesian inference of a new morphological dataset. We find a reduced membership of the mammalian stem-lineage, which excludes varanopids. This implies that evolutionary turnover of the mammalian stem-lineage during the early/middle Permian transition (273 Ma) was more abrupt than has previously been recognised. We also find that Parareptilia are nested within Diapsida. This suggests that temporal fenestration, a key structural innovation with important functional implications, evolved fewer times than generally thought, but showed highly variable morphology among early reptiles after its initial origin. Our phylogeny also addresses controversiesover the affinities of mesosaurids, the earliest known aquatic amniotes,which we recover as early-diverging parareptiles.