We report the computational investigation of the molecular geometries of a pair of manganese(III) spin crossover complexes. For the high-spin geometry, the density functionals significantly overestimate the Mn−Namine bond distances, although the geometry for the intermediate-spin is well-described. Comparisons with several wavefunction-based methods demonstrate that this error is due to the limited ability of density functional theory (DFT) to recover dispersion beyond a certain extent. Among the methods employed for geometry optimization, Møller-Plesset perturbation theory (MP2) appropriately describes the high-spin geometry, but results in a slightly reduced Mn−O distance in both the spin-states. On the other hand, complete active space second-order perturbation theory (CASPT2) results in a good description of the geometry for the intermediate spin state, but also sufficiently recovers dispersion performing well for the high-spin state. Despite the fact that the electronic structure of both spin states is dominated by one electron configuration, CASPT2 offers a balanced approach leading to molecular geometries with much better accuracy than MP2 and DFT. A scan along the Mn−Namine bond demonstrates that coupled cluster methods (i.e., DLPNO-CCSD(T)) also yield bond distances in agreement with experiment, while multiconfiguration pair density functional theory (MC-PDFT) is unable to recover dispersion well enough, analogous to single reference DFT.