For the prediction of a protein's function from its 3D-structure alone, it is of importance to elucidate by which properties functional site residues in a protein are discriminated from other residues. Here, we calculated five kinds of geometrical or physical properties of each residue in a protein. Those properties were integrated with techniques of multivariate analysis such as principal component analysis (PCA) or kernel PCA. Consequently, functional residues were found to show some distinct distributions in the scatter plot of those integrated data, which led to the proposal of a method for functional site prediction with a good performance. (C) 2010 Elsevier B.V. All rights reserved.