Purine nucleoside phosphorylase (PNP) catalyzes the reversible phosphorolysis of purine ribonucleosides and 2'-deoxyribonucleosides, yielding the purine base and (2'-deoxy)ribose 1-phosphate as products. While this enzyme has been extensively studied, several questions with respect to the catalytic mechanism have remained largely unanswered. The role of the phosphate and key amino acid residues in the catalytic reaction as well as the purine ring protonation state is elucidated using density functional theory calculations and extensive empirical valence bond (EVB) simulations. Free energy surfaces for adenosine, inosine, and guanosine are fitted to ab initio data and yield quantitative agreement with experimental data when the surfaces are used to model the corresponding enzymatic reactions. The cognate substrates 6-aminopurines (inosine and guanosine) interact with PNP through extensive hydrogen bonding, but the substrate specificity is found to be a direct result of the electrostatic preorganization energy along the reaction coordinate. Asn243 has previously been identified as a key residue providing substrate specificity. Mutation of Asn243 to Asp has dramatic effects on the substrate specificity, making 6-amino- and 6-oxopurines equally good as substrates. The principal effect of this particular mutation is the change in the electrostatic preorganization energy between the native enzyme and the Asn243Asp mutant, clearly favoring adenosine over inosine and guanosine. Thus, the EVB simulations show that this particular mutation affects the electrostatic preorganization of the active site, which in turn can explain the substrate specificity.