Single capillaries in the mesenteries of pithed frogs were perfused sequentially with two frog Ringer solutions. The first solution contained no protein; the second solution contained either native or chemically modified bovine serum albumin (BSA) at a concentration of 3-5 mg ml-1. During each perfusion capillary permeability was assessed from the hydraulic conductivity of the capillary wall (Lp) which was determined from measurements of fluid filtration rate at two or more different capillary pressures (Michel, Mason, Curry, Tooke & Hunter, 1974). Lp measured during perfusion with protein-free Ringer solution was on average three times greater than its value for the same vessel perfused with Ringer solution containing native BSA. This confirms the findings of Mason, Curry & Michel (1977). BSA, which had been succinylated to modify the free amino groups of its lysine residues, appeared to be as effective as native BSA in reducing Lp. After modification of its arginine side chains by exposure to 1,2-cyclohexanedione (CHD) in the presence of 0.2 M-NaOH, BSA lost its property of reducing Lp in capillaries perfused with Ringer solution. Exposure of BSA to 0.2 M-NaOH followed by dialysis against normal Ringer solution did not affect its property of reducing Lp. CHD-treated BSA at a concentration of 2.5 mg ml-1 had no effect upon the effective osmotic pressure exerted across capillary walls by Ringer perfusates containing the neutral polymer Ficoll 70 at a concentration of 40 mg ml-1. Native BSA raised the effective osmotic pressure from 7.07 +/- 1.93 cmH2O to 20.50 +/- 2.37 cmH2O (n = 7; P less than 0.001). It is concluded that the effects of BSA on permeability depend upon specific sites in the BSA molecule. It is suggested that these sites involve positively charged arginine side chains of the albumin molecule. The results are discussed in terms of the fibre-matrix hypothesis of capillary permeability and in terms of Brown's (1976) theory for the structure of albumin.