Local density approximation (LDA) and Becke88/Perdew86 gradient-corrected density functional theory calculations are used to estimate the heterolytic bond energy, E(M-X), corresponding to the process [MX6](g)(2-) --> M-g(4+) + 6X(g)(-). The computed data, including scalar relativistic corrections for second- and third-row metals, are benchmarked against updated values for the bonds Zr-Cl, Mo-Cl, Pd-CI, Sn-Cl, Hf-Cl, W-Cl, W-Br, Re-Cl, Re-Br, Os-Cl, Ir-CI, Pt-CI, Pt-Br, Ti-Cl, Ti-Br, and Ni-F derived from a combination of thermochemical and computational data on the antifluorite A(2)MX(6) hexahalometallate(IV) salts. The LDA tends to overbind, and the bond energies are generally too large. The BP method systematically reduces these values by about 60 kJ mol(-1), giving a significantly better comparison with experiment. However, LDA-optimized M-X bond lengths, both in vacuo and for a model ’in crystal’ {K-8[PdCl6]}(6+) cluster, are generally in better agreement with experiment.