High-level ab initio calculations with a variety of G2-based methods have been used to determine the structures and heats of formation of the alkali and alkaline earth oxides and hydroxides (M2O, MOH with M = Li, Na, and K; M'O, M'(OH)(2) with M' = Be, Mg, and Ca). Standard G2 theory, which is normally very reliable for the prediction of molecular thermochemistry, is shown to be quite unsuitable for the prediction of the heats of formation of several of these highly polar species, with errors greater than 100 kJ mol(-1) in some cases. Our calculations confirm that for systems containing the third-row atoms K and Ca, it is essential to include the 3s and 3p orbitals in the correlation space; Interestingly, an analogous relaxed-inner-valence (denoted riv) procedure is more beneficial for the Li- and Be-containing oxides and hydroxides than for the Na- and Mg-containing molecules. Inclusion of all orbitals in the correlation space (denoted full) generally provides only a slight further improvement, to the results. Removal of the additivity approximation of standard G2 theory through direct large basis set QCISD(T) calculations [e.g., G2(dir,full)] has a large effect for the oxides CaO and K2O. The QCISD(T) component of the G2 energy is poorly described for CaO, Na2O, and K2O, but this can be rectified through replacement of QCISD(T) with CCSD(T) [e.g., G2[CC](dir,full)], For five molecules (CaO, Be(OH)(2), Mg(OH)(2), Ca(OH)(2), and K2O) where significant discrepancies (10-30 kJ mol(-1)) remain between the best theoretical heats of formation (i.e., G2[CC](dir,full)) and experimental values, we suggest that experimental reexamination is desirable. Structures determined at the MP2/6-311fG(3df,2p) level are in good agreement with available experimental data. Structures obtained at the standard MP2/631G(d) level of G2 theory are not as good, but the impact of using the simpler geometries on calculated heats of formation is generally relatively small.