We demonstrate the utility of normal mode analysis in correctly predicting the binding modes of inhibitors in the active sites of matrix metalloproteinases (MMPs). We show the accuracy in predicting the positions of MMP-3 inhibitors is strongly dependent on which structure is used as the target, especially when it has been energy minimized. This dependency can be overcome by using intermediate structures generated along one of the normal modes previously calculated for a given target. These results may be of prime importance for further in silico drug discovery.