Knowledge of the protonation states of the ionizable residues in an enzyme is a prerequisite to an accurate description of its structure and mechanism. In practice, the use of the inappropriate protonation state for an amino acid in a molecular modeling computation (e.g., molecular dynamics simulation) is likely to lead to unrealistic results. Although methods using solvers of the linearized Poisson-Boltzmann equation have proven to yield accurate pK(a) predictions, they bear a number of limitations. They are quite demanding in terms of computational power and are sensitive to representation of the charges and their position (force field and protein conformation). Moreover they depend on the choice of a dielectric constant for the protein interior. In this manuscript, we describe the difficulties met when trying to predict the protonation state of a buried amino acid, located in a protein for which very little biochemical data is available. Such a case is highly representative of the challenges faced in theoretical biology studies. Proteinase 3 (PR3) is an enzyme involved in proteolytic events associated with inflammation. It is a potential target in the development of new anti-inflammatory therapeutic strategies. We report the results of pK(a) predictions of the aspartic acid 213 of PR3 with a FDPB solver. We probed the influence of the choice of the dielectric constant for the protein interior epsilon(p) and the benefits of conformational sampling by molecular dynamics (MD) on the pK(a) prediction of this carboxylate group. Using only the FDPB calculations, we could not conclude on the protonation state of Asp213. MD simulations confronted to knowledge of the ligand-binding and reaction mechanism led us to decide on a protonated form of this aspartic acid. We also demonstrate that the use of the wrong protonation state leads to an unreliable structural model for PR3. pK(a) prediction with a fast empirical method yielded a pK(a) of 8.4 for Asp213, which is in agreement with our choice of protonation state based on MD simulations.