The intravenous injection into mice of small volumes (less than 0.1 ml) of peroxidatic enzymes of molecular weight of 40 000 daltons or greater results in little if any penetration of these probe molecules into endothelial junctions. The injection of cytochrome c (12 000 daltons), on the other hand, results in the localization of this tracer in some but not all endothelial junctions. When horseradish peroxidase (EC 1.11.1.7) is injected in a large volume of saline (0.5 ml), reaction product is present in endothelial junctions and basement membrane, but is prevented from entering the alveolar space by zonulae occludentes between epithelial cells. These experiments indicate that although endothelial junctions, under physiological conditions, are largely impermeable to molecules the size of horseradish peroxidase, and presumably most serum proteins, they are labile and susceptible to stretching if intravascular pressure is increased. Freeze-fracture studies show that pulmonary capillary endothelial junctions are composed of one or at the most two strands which show areas of discontinuity. Epithelial junctions, by contrast, are composed of a continuous, complex network of anastomosing fibres. These observations confirm physiological experiments which indicate that it is the pulmonary epithelium rather than the endothelium which determines the permeability properties of the alveolar-capillary membrane to lipid-insoluble molecules. Bidirectional pinocytic transport is an additional mechanism whereby lipid-insoluble molecules are transported across both endothelial and epithelial layers. The relative contribution of this transport mechanism to the total amount transported remains to be established.