Rational design and engineering of membrane-active peptides remains a largely unsatisfied goal. We have hypothesized that this is due, in part, to the fact that some membrane activities, such as permeabilization, are not dependent on specific amino acid sequences or specific three-dimensional peptide structures. Instead they depend on interfacial activity: the ability of a molecule to partition into the membrane-water interface and to alter the packing and organization of lipids. Here we test that idea by taking a nonclassical approach to biomolecular engineering and design of membrane-active peptides. A 16,384-member rational combinatorial peptide library, containing peptides of 9-15 amino acids in length, was screened for soluble members that permeabilize phospholipid membranes. A stringent, two-phase, high-throughput screen was used to identify 10 unique peptides that had potent membrane-permeabilizing activity but were also water soluble. These rare and uniquely active peptides do not share any particular sequence motif, peptide length, or net charge, but instead they share common compositional features, secondary structure, and core hydrophobicity. We show that they function by a common mechanism that depends mostly on interfacial activity and leads to transient pore formation. We demonstrate here that composition-space peptide libraries coupled with function-based high-throughput screens can lead to the discovery of diverse, soluble, and highly potent membrane-permeabilizing peptides.