Proteochemometrics is a technology for the study of molecular recognition based on chemometric techniques. Here we applied it to analyse the amino acids and amino acid physico-chemical properties that are involved in antibodies' recognition of peptide antigens. To this end, we used a study system comprised by a diverse single chain antibody library derived from the murine mAb anti-p24 (HIV-1) antibody CB4-1, evaluated on peptide arrays manufactured by SPOT synthesis. The binding pattern obtained was correlated to physico-chemical descriptors (z-scales) of antibodies and peptides amino acids using partial least-squares projections to latent structures. Cross terms derived from antibody and antigen descriptors were included, which substantially improved the proteochemometric model. The final model was statistically highly satisfactory with a correlation coefficient R(2) = 0.73 and predictive ability Q(2) = 0.68. The physico-chemical properties of each interacting amino acid residue of both the peptides and the antibodies being essential for the antigen-antibody recognition could be retrieved from the model. The study shows for the first time the feasibility of using proteochemometrics to analyse the molecular recognition of antigens by antibodies.