A set of 41 metal-ligand bond distances in 25 third-row transition-metal complexes, for which precise structural data are known in the gas phase, is used to assess optimized and zero-point averaged geometries obtained from DFT computations with various exchange-correlation functionals and basis sets. For a given functional (except LSDA) Stuttgart-type quasi-relativistic effective core potentials and an all-electron scalar relativistic approach (ZORA) tend to produce very similar geometries. In contrast to the lighter congeners, LSDA affords reasonably accurate geometries of 5d-metal complexes, as it is among the functionals with the lowest mean and standard deviations from experiment. For this set the ranking of some other popular density functionals, ordered according to decreasing standard deviation, is BLYP > VSXC > BP86 ≈ BPW91 ≈ TPSS ≈ B3LYP ≈ PBE > TPSSh > B3PW91 ≈ B3P86 ≈ PBE hybrid. In this case hybrid functionals are superior to their nonhybrid variants. In addition, we have reinvestigated the previous test sets for 3d- (Bühl M.; Kabrede, H. J. Chem. Theory Comput. 2006, 2, 1282-1290) and 4d- (Waller, M. P.; Bühl, M. J. Comput. Chem. 2007, 28, 1531-1537) transition-metal complexes using all-electron scalar relativistic DFT calculations in addition to the published nonrelativistic and ECP results. For this combined test set comprising first-, second-, and third-row metal complexes, B3P86 and PBE hybrid are indicated to perform best. A remarkably consistent standard deviation of around 2 pm in metal-ligand bond distances is achieved over the entire set of d-block elements.