Pentadecapeptides based on modified murine lactoferricin (LFM) sequences show varying degrees of antibacterial activity against Escherichia coli and Staphylococcus aureus. By means of projections to latent structures (PLS), a good correlation is obtained if the biological activity is modelled as a function of variables describing peptide properties, e.g. alpha-helicity, hydrophobicity/hydrophilicity and charge. Using variables derived from a principal component analysis (PCA) of all naturally occurring amino acids, it is possible to describe the amino acid content of the peptides using three variables per amino acid position. The resulting descriptor matrix is then used to develop quantitative structure-activity relationships (QSAR). It is shown that the theoretically derived descriptors model the activity of the peptides better than the earlier model, and that properties of the peptides other than antibacterial activity can be predicted.