This work presents a detailed evaluation of the performance of density functional theory (DFT) for the prediction of zero-field splittings (ZFSs) in Mn(II) coordination complexes. Eighteen experimentally well characterized four-, five-, and six-coordinate complexes of the general formula [Mn(L),L'(2)] with L' = Cl, Br, I, NCS, or N-3 (L = an oligodentate ligand) are considered. Several DFT-based approaches for the prediction of the ZFSs are compared. For the estimation of the spin-orbit coupling (SOC) part of the ZFS, it was found that the Pederson-Khanna (PK) approach is more successful than the previously proposed quasi-restricted orbitals (QRO)-based method. In either case, accounting for the spin-spin (SS) interaction either with or without the inclusion of the spin-polarization effects improves the results. This argues for the physical necessity of accounting for this important contribution to the ZFS. On average, the SS contribution represents similar to 30% of the axial D parameters. In addition to the SS part, the SOC contributions of d-d spin flip (alpha beta) and ligand-to-metal charge transfer excited states 00) were found to dominate the SOC part of the D parameter; the observed near cancellation between the alpha alpha and beta alpha parts is discussed in the framework of the PK model. The calculations systematically (correlation coefficient similar to 0.99) overestimate the experimental D values by similar to 60%. Comparison of the signs of calculated and measured D values shows that the signs of the calculated axial ZFS parameters are unreliable once EID > 0.2. Finally, we find that the calculated D and EID values are highly sensitive to small structural changes. It is observed that the use of theoretically optimized geometries leads to a significant deterioration of the theoretical predictions relative to the experimental geometries derived from X-ray diffraction. The standard deviation of the theoretical predictions for the D values almost doubles from similar to 0.1 to similar to 0.2 cm(-1) upon using quantum chemically optimized structures. We do not find any noticeable improvement in considering basis sets larger than standard double-(SVP) or triple-zeta (TZVP) basis sets or using functionals other than the BP functional.