This paper introduces an economical approach for improving the accuracy and convergence of quantum mechanics/molecular mechanics (QM/MM) models. The approach is tested on a series of neutral and charged amino acids embedded in a 160-water cluster, where their intramolecular proton transfer energies (neutral amino acid → zwitterionic amino acid) were previously obtained at the ωB97X-D/6-31G(d) level of theory. When the charges on the MM atoms were replaced with those obtained at the same QM level of theory used to treat the QM atoms, this significantly improved the accuracy and convergence of the QM/MM models. In particular, the QM/MM model converged to within 1.4 kcal mol-1 of directly calculated DFT energies for smaller (by as many as 20 waters) QM regions. The use of atomic charges obtained from the natural population analysis yielded the most significant improvement, while other charge schemes such as Mulliken, electrostatic potential, or CM5 led to poorer outcomes. It is further demonstrated that the QM atomic charges can be accurately estimated in a highly efficient manner using an iterative fragmentation approach based on the moving-domain QM/MM method. Similar observations were made when the approach was used to predict the barrier of an SN2 reaction. Thus, the use of QM-quality atomic charges on MM atoms represents a simple and easy-to-implement strategy for improving the accuracy of QM/MM models.