Capillary leak in severe sepsis involves disruption of endothelial cell tight junctions. We modeled this process by TNF treatment of cultured human dermal microvascular endothelial cell (HDMEC) monolayers, which unlike human umbilical vein endothelial cells form claudin-5-dependent tight junctions and a high-resistance permeability barrier. Continuous monitoring with electrical cell-substrate impedance sensing revealed that TNF disrupts tight junction-dependent HDMEC barriers in discrete steps: an ~5% increase in transendothelial electrical resistance over 40 minutes; a decrease to ~10% below basal levels over 2 hours (phase 1 leak); an interphase plateau of 1 hour; and a major fall in transendothelial electrical resistance to < 70% of basal levels by 8-10 hours (phase 2 leak), with EC50 values of TNF for phase 1 and 2 leak of ~30 and ~150 pg/ml, respectively. TNF leak is reversible and independent of cell death. Leak correlates with disruption of continuous claudin-5 immunofluorescence staining, myosin light chain phosphorylation and loss of claudin-5 co-localization with cortical actin. All these responses require NF-κB signaling, shown by inhibition with Bay 11 or overexpression of IκB super-repressor, and are blocked by H-1152 or Y-27632, selective inhibitors of Rho-associated kinase that do not block other NF-κB-dependent responses. siRNA combined knockdown of Rho-associated kinase-1 and -2 also prevents myosin light chain phosphorylation, loss of claudin-5/actin co-localization, claudin-5 reorganization and reduces phase 1 leak. However, unlike H-1152 and Y-27632, combined Rho-associated kinase-1/2 siRNA knockdown does not reduce the magnitude of phase 2 leak, suggesting that H-1152 and Y-27632 have targets beyond Rho-associated kinases that regulate endothelial barrier function. We conclude that TNF disrupts TJs in HDMECs in two distinct NF-κB-dependent steps, the first involving Rho-associated kinase and the second likely to involve an as yet unidentified but structurally related protein kinase(s).