The performance of computationally accessible levels of calculation for the transition states of organocatalytic reaction has been assessed. Reference post-Hartree–Fock single point energy calculations were used as standards for the gas-phase Born–Oppenheimer relative energies of pairs of alternative transition states that lead to the two product enantiomers. We show that semiempirical methods cannot even be relied on to yield qualitatively correct results. The geometries (optimized, for instance, with DFT) have a large impact on the results of high-level post-HF calculations, so that it is essential to use an adequate DFT technique and basis set. DFT can yield quantitatively correct results that are consistent with post-HF calculations if functionals that consider dispersion are used. Geometries for large systems show larger errors than those for smaller ones but are treated better by functionals such as M06-2X and w97Bxd that include dispersion implicitly or explicitly. Local correlation techniques introduce errors of comparable magnitude to those given by different levels of geometry optimization. We recommend RICC2/TZVP//M06-2X/TZVP, RI-MP2/TZVP// M06-2X/TZVP, and M06-2X/TZVP// M06-2X/TZVP calculations in that order, depending on the size of the system.