In designing mutagenesis experiments, it is often crucial to know how certain mutations will affect the structure and thermodynamic stability of the protein. Here, we present a methodology, Eris, to efficiently and accurately compute the stability changes of proteins upon mutations using our protein-modeling suite, Medusa. We evaluate the stability changes upon mutations for 595 mutants from five structurally unrelated proteins, and find significant correlations between the predicted and experimental results. For cases when the high-resolution protein structure is not available, we find that better predictions are obtained by backbone structure prerelaxation. The advantage of our approach is that it is based on physical descriptions of atomic interactions, and does not rely on parameter training with available experimental protein stability data. Unlike other methods, Eris also models the backbone flexibility, thereby allowing for determination of the mutation-induced backbone conformational changes. Eris is freely available via the web server at http://eris.dokhlab.org.