How accurate are DFT treatments of organic energies?

Increasing awareness that popular functionals fail to describe many energies accurately has ended expectations of black-box DFT usage. The performance of nine density functionals, compared by computing the bond separation energies of 72 illustrative hydrocarbons with available experimental data, reveals that only Zhao and Truhlar's recently proposed M05-2X functional, with a 2.13 kcal/mol average deviation from experiment, performs satisfactorily. B3LYP and other functionals show larger deviations.