Abstract: Correctly understanding and predicting binding site residues is necessary for studying protein interactions and their networks. A novel approach was reported to predict protein interface residues using five dimensional feature space based on residue physical and chemical features. In this method, a grid multiple dimensional feature space was built using standardized feature values. Then, all the surface residues were put into the grids according to their feature values. Finally, the grids were clustered. Interestingly, interface residues prefer some grids clustered together. Excellent prediction result were obtained on a public and data benchmark was verified. This approach not only opens up a new visual angle for binding site residue prediction, but also help to understand protein-protein interactions more deeply.