The major histocompatibility complex (MHC)Ⅰ binding affinity models have contributed to screen the candidate peptides, and have assisted the experiments in determining the peptides that can form complexes with MHCⅠmolecules to activate cytotoxic T cells. The transporter associated with antigen processing (TAP) binding models could also be used for screening the candidate peptides. How to make the best of the two types of binding affinity models for screening out the candidate peptides, the similarities and differences between the selectivity of TAP and MHCⅠmolecules in peptides and the biological mechanism of that similarities and differences, these three questions remains obscure. Herein, we rearranged the TAP binding test set, increasing its size to 699. The established TAP binding model based on Kernel-function stabilized matrix method (KSMM) had a higher prediction accuracy than that of the state of the art, achieving a relevant correlation coefficient of 0.89 on a 5-fold cross-validation. The integrative prediction of HLA-A3 affinity and TAP affinity models remarkedly improved the discrimination accuracy, with the AUC value increasing from ~0.82 to 0.87. This improvement is due to the different preferences of the two types of affinity models for the best defined amino acids on 2nd and 9th positions, as well as to the complementarity of such different preferences. The results of TAP-peptide docking also supported this conclusion. The TAP model is available online: http://www.bilologymaths.top/mbtwo/major.aspx.