An improved statistical-acoustics model of high-frequency sound fields in coupled rooms is developed by incorporating into prior models geometrical-acoustics corrections for both energy decay within subrooms and energy transfer between subrooms. The conditions under which statistical-acoustics models of coupled rooms are valid approximations to geometrical acoustics are examined by comparison of computational geometrical-acoustics predictions of decay curves in two- and three-room systems with those of both improved and prior statistical-acoustics models. The accuracy of the decay model used within subrooms is found to have a primary influence on the accuracy of predictions in coupled systems. Likewise, nondiffuse transfer of energy is shown to significantly affect decay of energy in systems of coupled rooms. The decrease in energy density of the reverberant field with distance from the source, which is predicted by geometrical acoustics, is found to result in spatial dependence of decay-curve shape for certain coupling geometries. Geometrical effects are shown to contribute to the failure of statistical-acoustics models in the case of strong coupling between subrooms; thus, previously proposed statistical-acoustics criteria cannot predict the point at which the models break down with consistent accuracy.