How accurate is the CASPT2 method?

This paper is a reply to a recent paper on the isomers of trisulfur oxides (Chem. Commun., 2005, 3712) where the accuracy of the CASPT2 method is called in question. We illustrate that their results are due to an inadequate choice of active orbitals. Some results for related O and S containing molecules are also presented to strengthen the argumentation. It is concluded that the CASPT2 method is accurate to at least 0.1 eV (2 kcal mol(-1)) for this type of systems.