The oxygen-atom transfer reaction catalyzed by the mononuclear molybdenum enzyme dimethyl sulfoxide reductase (DMSOR) has attracted considerable attention by both experimental and theoretical studies. We show here that this reaction is more sensitive to details of quantum mechanical calculations that what has previously been appreciated. Basis sets of at least triplezeta quality are needed to obtain qualitatively correct results. Dispersion has an appreciable effect on the reaction, in particular the binding of the substrate or the dissociation of the product (up to 34 kJ/mol). Polar and non-polar solvation effects are also significant, especially if the enzyme can avoid cavitation effects by using a pre-formed active-site cavity. Relativistic effects are considerable (up to 22 kJ/mol), but they are reasonably well treated by a relativistic effective core potential. Various density-functional methods give widely different results for the activation and reaction energy (differences of over 100 kJ/mol), mainly reflecting the amount of exact exchange in the functional, owing to the oxidation of Mo from +IV to +VI. By calibration towards local CCSD(T0) calculations, we show that none of eight tested functionals (TPSS, BP86, BLYP, B97-D, TPSSH, B3LYP, PBE0, and BHLYP) give accurate energies for all states in the reaction. Instead, B3LYP gives the best activation barrier, whereas pure functionals give more accurate energies for the other states. Our best results indicate that the enzyme follows a two-step associative reaction mechanism with an overall activation enthalpy of 63 kJ/mol, which is in excellent agreement with the experimental results.