WW domains are small globular protein interaction modules found in a wide spectrum of proteins. They recognize their target proteins by binding specifically to short linear peptide motifs that are often proline-rich. To infer the determinants of the ligand binding propensities of WW domains, we analyzed 42 WW domains. We built models of the 3D structures of the WW domains and their peptide complexes by comparative modeling supplemented with experimental data from peptide library screens. The models provide new insights into the orientation and position of the peptide in structures of WW domain-peptide complexes that have not yet been determined experimentally. From a protein interaction property similarity analysis (PIPSA) of the WW domain structures, we show that electrostatic potential is a distinguishing feature of WW domains and we propose a structure-based classification of WW domains that expands the existent ligand-based classification scheme. Application of the comparative molecular field analysis (CoMFA), GRID/GOLPE and comparative binding energy (COMBINE) analysis methods permitted the derivation of quantitative structure-activity relationships (QSARs) that aid in identifying the specificity-determining residues within WW domains and their ligand-recognition motifs. Using these QSARs, a new group-specific sequence feature of WW domains that target arginine-containing peptides was identified. Finally, the QSAR models were applied to the design of a peptide to bind with greater affinity than the known binding peptide sequences of the yRSP5-1 WW domain. The prediction was verified experimentally, providing validation of the QSAR models and demonstrating the possibility of rationally improving peptide affinity for WW domains. The QSAR models may also be applied to the prediction of the specificity of WW domains with uncharacterized ligand-binding properties.