This letter reports the apparent failure of a nominally high-accuracy ab initio calculation of a modified G2 type for a reaction that is of considerable importance in hydrocarbon combustion. The failure is traced to the reliance on single-reference methods in the treatment of electron correlation. Its occurrence serves as a general caution about the use of such calculations for mechanistic analysis.