Defensins comprise a family of broad-spectrum antimicrobial peptides that are stored in the cytoplasmic granules of mammalian neutrophils and Paneth cells of the small intestine. Neutrophil defensins are known to permeabilize cell membranes of susceptible microorganisms, but the mechanism of permeabilization is uncertain. We report here the results of an investigation of the mechanism by which HNP-2, one of 4 human neutrophil defensins, permeabilizes large unilamellar vesicles formed from the anionic lipid palmitoyloleoylphosphatidylglycerol (POPG). As observed by others, we find that HNP-2 (net charge = +3) cannot bind to vesicles formed from neutral lipids. The binding of HNP-2 to vesicles containing varying amounts of POPG and neutral (zwitterionic) palmitoyloleoylphosphatidylcholine (POPC) demonstrates that binding is initiated through electrostatic interactions.Because vesicle aggregation and fusion can confound studies of the interaction of HNP-2 with vesicles, those processes were explored systematically by varying the concentrations of vesicles and HNP-2, and the P0PG:POPC ratio. Vesicles (300 pM POPG) readily aggregated at HNP-2 concentrations above 1 pM, but no mixing of vesicle contents could be detected for concentrations as high as 2 pM despite the fact that intervesicular lipid mixing could be demonstrated. This indicates that if fusion of vesicles occurs, it is hemi-fusion, in which only the outer monolayers mix at bilayer contact sites. Under conditions of limited aggregation and intervesicular lipid mixing, the fractional leakage of small solutes is a sigmoidal function of peptide concentration. For 300 pM POPG vesicles, 50% of entrapped solute is released by 0.7 p M HNP-2. We introduce a simple method for determining whether leakage from vesicles is graded or all-or-none. We show by means of this fluorescence "requenching" method that native HNP-2 induces vesicle leakage in an all-or-none manner, whereas reduced HNP-2 induces partial, or graded, leakage of vesicle contents. At HNP-2 concentrations that release 100% of small (-400 Da) markers, a fluorescent dextran of 4,400 Da is partially retained in the vesicles, and a 18,900-Da dextran is mostly retained. These results suggest that HNP-2 can form pores that have a maximum diameter of approximately 25 A. A speculative multimeric model of the pore is presented based on these results and on the crystal structure of a human defensin.