The tight junction (zonula occludens) links epithelial cells into a monolayer by forming a continuous belt of sealing contacts around the apex of each cell. They appear in thin sections as if they were ‘fusions’ between the apposed plasma membranes¹ and in freeze-fracture replicas as patterns of complementary strands and furrows². These images have led to the proposal that the core of the tight junction is formed by a hexagonal cylinder of lipids^3,4. In this model, the cytoplasmic leaflet of the apical and basolateral plasma membrane domains would be continuous, whereas the exoplasmic leaflets of the two plasma membrane domains of the same cell would be separated at the tight junction and are instead predicted to be continuous between the plasma membranes of neighbouring cells. We demonstrate here that this prediction does not hold true. An endogenous glycolipid (Forssman antigen), present in the exoplasmic leaflet of the apical membrane of MDCK strain II cells^5,6, is unable to pass to MDCK strain I cells (which lack this glycolipid) under conditions where these cells are connected by tight junctions. In addition, fluorescent lipids which have been fused into the plasma membrane^7,8 of one MDCK cell do not diffuse to neighbouring cells while the tight junctions between the cells are intact.