A new structure-based approach was proposed to quantitatively characterize the binding profile of human amphiphysin-1 (hAmph1) SH3 domain-peptide complexes. In this protocol, the protein/peptide atoms were classified into 16 types in terms of their physicochemical meaning and biological function, and then a 16 x 16 atom-pair interaction matrix was constructed to describe 256 atom-pair types between the SH3 domain and the peptide ligand, with atoms from peptide and SH3 domain served as the matrix columns and rows, respectively. Three non-covalent effects dominating SH3 domain-peptide binding as electrostatic, van der Waals (steric) and hydrophobic interactions were separately calculated for the 256 atom-pair types. As a result, 768 descriptors coding detailed information about SH3 domain-peptide interactions were yielded for further statistical modeling and analysis. Based on a culled data set consisting of 592 samples with known affinities, we employed this approach, coupled with partial least square (PLS) regression and genetic algorithm (GA), to predict and to interpret the peptide-binding behavior to SH3 domain. In comparison with the previous works, our method is more capable of capturing important factors in the SH3 domain-peptide binding, thus, yielding models with better statistical performance. Furthermore, the optimal GA/PLS model indicates that the electrostatic effect plays a crucial role in SH3 domain-peptide complexes, and steric contact and hydrophobic force also contribute significantly to the binding.