The performance of Gaussian-2 theory is investigated when higher level theoretical methods are included for correlation effects, geometries, and zero-point energies. A higher level of correlation treatment is examined using Brueckner doubles [BD(T)] and coupled cluster [CCSD(T)] methods rather than quadratic configuration interaction [QCISD(T)]. The use of geometries optimized at the QCISD level rather than the second-order Moller-Plesset level (MP2) and the use of scaled MP2 zero-point energies rather than scaled Hartree-Fock (HF) zero-point energies have also been examined. The set of 125 energies used for validation of G2 theory [J. Chem. Phys, 94, 7221 (1991)] is used to test out these variations of G2 theory. Inclusion of higher levels of correlation treatment has little effect except in the cases of multiply-bonded systems. In these cases better agreement is obtained in some cases and poorer agreement in others so that there is no improvement in overall performance. The use of QCISD geometries yields significantly better agreement with experiment for several cases including the ionization potentials of CS and O-2, electron affinity of CN, and dissociation energies of N-2, O-2, CN, and SO2. This leads to a slightly better agreement with experiment overall. The MP2 zero-point energies gives no overall improvement. These methods may be useful for specific systems.