We developed a new scoring method that selects a protein-ligand complex structure with higher geometrical accuracy than the top-scoring complex structure, using the structural information of known protein-ligand complexes. To apply this method, one or more protein-ligand complex structures must be known for the target protein. A number of predicted structures were generated by the protein-compound docking program for a new ligand, and one of these structures, which showed the maximum overlap with the ligand coordinates of the known protein-ligand complex, was selected as the most probable complex structure.