Severe influenza infections are complicated by acute lung injury, a syndrome of pulmonary microvascular leak. The pathogenesis of this complication is unclear. We hypothesized that human influenza could directly infect the lung microvascular endothelium, leading to loss of endothelial barrier function. We infected human lung microvascular endothelium with both clinical and laboratory strains of human influenza. Permeability of endothelial monolayers was assessed by spectrofluorimetry and by measurement of the transendothelial electrical resistance. We determined the molecular mechanisms of flu-induced endothelial permeability and developed a mouse model of severe influenza. We found that both clinical and laboratory strains of human influenza can infect and replicate in human pulmonary microvascular endothelium, leading to a marked increase in permeability. This was caused by apoptosis of the lung endothelium, since inhibition of caspases greatly attenuated influenza-induced endothelial leak. Remarkably, replication-deficient virus also caused a significant degree of endothelial permeability, despite displaying no cytotoxic effects to the endothelium. Instead, replication-deficient virus induced degradation of the tight junction protein claudin-5; the adherens junction protein VE-cadherin and the actin cytoskeleton were unaffected. Over-expression of claudin-5 was sufficient to prevent replication-deficient virus-induced permeability. The barrier-protective agent formoterol was able to markedly attenuate flu-induced leak in association with dose-dependent induction of claudin-5. Finally, mice infected with human influenza developed pulmonary edema that was abrogated by parenteral treatment with formoterol. Thus, we describe two distinct mechanisms by which human influenza can induce pulmonary microvascular leak. Our findings have implications for the pathogenesis and treatment of acute lung injury from severe influenza.