Many different types of organisms use antimicrobial peptides, typically 20-40 amino acids in length, for defence against infection. Most are capable of rapidly killing a wide range of microbial cells. They have been classified according to their active structures into six extensive groups. It is not yet clear how these peptides kill bacterial cells, but it is widely believed that some cationic antimicrobial peptides kill by disrupting bacterial membranes, allowing the free exchange of intra- and extra-cellular ions. The selectivity of these peptides appears to relate to differences between the external membranes of prokaryotic and eukaryotic cells. The action of the peptides may involve the formation of 'barrel-stave' or 'torroidal' pores, the introduction of packing defects in the membrane phospholipids, or large-scale disruption of the membrane by a very dense aggregation of parallel-oriented peptide, called the 'carpet mechanism'. Antimicrobial peptides are attractive candidates for clinical development because of their selectivity, their speed of action and because bacteria may not easily develop resistance against them. Some antimicrobial peptides are already in clinical and commercial use, including ambicin (nisin), polymixin B and gramicidin S. There have been several attempts at developing peptides to make them more suitable for clinical use. For those peptides that act against bacterial membranes, it is possible to differentiate between those structural features that contribute to the specificity of initial membrane binding and those that contribute to the subsequent breach of membrane integrity. The design of novel antimicrobial peptides would necessitate the optimisation of multiple parameters, a problem that has proved difficult to solve. Potential problems to be overcome include high production costs, toxicity against eukaryotic cells, susceptibility to proteolytic degradation and the development of allergies to the peptides. Biosynthesis, using recombinant DNA techniques, could make commercial-scale synthesis feasible but the peptides are usually lethal to the micro-organisms used to produce them. Proteolytic degradation can be reduced by modifying the peptides to contain nonstandard amino acids, or by restricting the use of peptides to topical applications. The problem of sensitisation could be overcome by the use of our own natural antibiotics to prevent or treat infections. Despite early hopes that bacteria would not easily develop resistance to antimicrobial peptides, it is clear that some strains of bacteria already have.