All-atoms molecular dynamics (MD) simulations are a very powerful tool to predict structural, dynamical, and thermodynamical properties of biological molecules and their complexes with ligands. Unfortunately, the current computational power constrains this analysis to time scales of ~100 ns, too short to follow several important biological processes, which evolve on a much longer time scale. To bridge the gap between time scales of feasible simulations and those of biologically relevant processes, several simplified methods (coarse-grained models) have been proposed in which a small number of interaction sites are used to represent the systems [1-3]. Unfortunately, since the chemical details are often neglected, these models are not suitable for investigating processes involving molecular recognition. Nevertheless, such processes are usually highly localized and involve small protein portions. A hybrid approach (MM/CG) has been developed in which a small portion of the protein (i.e. the binding pocket) is treated at the level of detail allowed by classical MD, while the remainder of the protein is treated at the CG level [4]. The method was tested on three proteins of great pharmacological relevance, belonging to the protease class (HIV-1 PR, human β-secretase and OmpT from E. coli) and was able to reproduce both the large-scale motions and the local features of the active sites of the considered proteins. Nevertheless, these simulations were performed in a few days, whereas the corresponding MD simulations would require months on the same computers. MM/CG approach emerges as a powerful tool to investigate long (in the order of μs) dynamics of enzymes, which may impact on function, as well as an efficient and fast tool for computational structural biology. By allowing running more numerous and longer simulations, MM/CG on one hand is expected to improve the confidence of the results and on the other, it may strengthen the interrelation between molecular biology experiments and simulations.