The replacement of a suboptimal amino acid in a primary anchor position with an optimal residue improves human leucocyte antigen (HLA) binding and immunogenicity, while maintaining cytotoxic T lymphocyte (CTL) specificity. Using a neural network capable of performing quantitative predictions of peptide binding to HLA-A2 molecules, we identified three p53 protein-derived nonamer peptides with intermediate binding owing to suboptimal amino acids in the P2 anchor position. These peptides were synthesized along with the corresponding analogs, where the natural P2 residue had been replaced with the optimal leucine residue. All three modified peptides bound to and more efficiently stabilized HLA-A2 molecules than the corresponding nonmodified peptides. The HLA-A2 transgenic mice were used for immunization. Two of the epitopes were more immunogenic in their modified than in their natural versions. The CTLs raised against the modified peptides efficiently killed the target cells pulsed with the corresponding native peptide. In terms of sensitizing the targets cells for the CTL killing, the modified peptides were more efficient than native peptides. Finally, the CTLs induced by modified peptide killed HLA-A2 transgenic mouse fibrosarcoma cells transfected with human p53 DNA. The data suggest that modified self-peptides derived from overexpressed tumour-associated proteins can be used in vaccine development against cancer, and that quantitative predictions of HLA binding is of value in the rational selection and improvement of target epitopes recognized by CTLs.