The elastic network model (ENM) is an effective method to extract the intrinsic dynamical properties encoded in protein tertiary structures. We have proposed a new ENM-based analysis method to reveal the motion modes directly responsible for a specific protein function, in which an internal coordinate related to the specific function was introduced to construct the internal/Cartesian hybrid coordinate space. In the present work, the function-related internal coordinates combined with a linear perturbation method were applied to identify the key sites controlling specific protein functional motions. The change in the fluctuations of the internal coordinate in response to residue perturbation was calculated in the hybrid coordinate space by using the linear response theory. The residues with the large fluctuation changes were identified to be the key sites that allosterically control the specific protein function. Two proteins, i.e., human DNA polymerase β and the chaperonin from Methanococcus maripaludis, were investigated as case studies, in which several collective and local internal coordinates were applied to identify the functionally key residues of these two studied proteins. The calculation results are consistent with the experimental observations. It is found that different collective internal coordinates lead to similar results, where the predicted functionally key sites are located at similar positions in the protein structure. While for the local internal coordinates, the predicted key sites tend to be situated at the region near to the coordinate-involving residues. Our studies provide a starting point for further exploring other function-related internal coordinates for other interesting proteins.