Abstract: The present invention relates to the identification of CTL epitopes by the combination of biochemical assays, statistical matrix calculations, and artificial neural networks. A set of peptide libraries are used to generate complete unbiased matrices representing peptide-MHC interactions used to generate a primary prediction of MHC binding for all possible non-redundant peptides. The best binders are subject to a quantitative biochemical binding assay and subsequently a computerised artificial neural network prediction program built from these in vitro experimental MHC-I binding data. The method further comprises improving the identified epitope by replacing amino acids, and testing the identified CTL epitopes in in vitro and in vivo models.