H2O2, is an attractive oxidant for synthetic chemistry, especially if activated as percarboxylic acid. H2O2, however, is also a potent inactivator of enzymes. Protein engineering efforts to improve enzyme resistance against H2O2 in the past have mostly focused on tedious probabilistic directed evolution approaches. Here we demonstrate that a rational approach combining multiscale MD simulations and Born–Oppenheimer ab initio QM/MM MD simulations is an efficient approach to rapidly identify improved enzyme variants. Thus, the lipase from Penicillium camembertii was redesigned with a single mutation (I260R), leading to drastic improvements in H2O2 resistance while maintaining the catalytic activity. Also the extension of this methodology to other enzymes is demonstrated.