Ab initio calculations have been performed at correlated levels of theory on several hydrocarbons and carbocations in order to investigate the basis set dependence of their computed hydrogenolysis energies, which determine the stabilities of these species relative to H-2 and CH4, and from which proton affinities, hydride ion affinities, and isomerization energies may be derived. One series of basis sets studied is derived from the 6-31G(d,p) basis by augmentation with diffuse functions on carbon and additional first and second polarization functions on carbon and hydrogen. Also studied, to a more limited extent, is the triple-split 6-311G(2df,2pd) basis. The 6-31G and 6-311G series of basis sets are compared with the correlation-consistent polarized valence double-, triple-, and quadruple-zeta basis sets of Dunning, in which the sp part of the basis is systematically improved as the polarization space is augmented. The hydrogenolysis energies computed with the 6-3lG basis sets show poorer convergence patterns upon augmentation, including divergence from experimental values in some cases, and show an excessive dependence on the presence of diffuse functions on carbon and on the choice of five versus six components for d polarization functions. These problems are attributed to an inadequate representation of the sp part in the valence double-split 6-31G basis sets. Better results, and reduced sensitivity with respect to the number of d components and to the addition of diffuse functions, are found with the triple-split 6-311G(2df,2pd) basis set. The correlation-consistent basis sets have been found to show smooth convergence of the computed hydrogenolysis energies, both internally and with respect to experimental data, and no significant dependence on the presence of diffuse functions or on the number of components for the d functions, except for the double-split member of the series. The correlation-consistent polarized valence triple-zeta basis set is recommended for accurate calculations of reaction energies, with 6-311 + G(2df,2pd) (with diffuse functions) a reasonable alternative.