Recently we presented an extension of the direct reaction field (DRF) method, in which a quantum system and a set of point charges and interacting polarizabilities are embedded in a continuum that is characterized by a dielectric constant ϵ and a finite ionic strength. The reaction field of the continuum is found by solving the (linearized) Poisson–Boltzmann equation by a boundary element method for the complete charge distribution in a cavity of arbitrary size and form. Like many other authors, we found that the results depend critically on the choice of the size of the cavity, in the sense that the continuum contribution to the solvation energy decreases rapidly with the relative cavity size. The literature gives no clues for the definition of the cavity size beyond “physical intuition” or implicit fitting to experimental or otherwise desired results. From theoretical considerations, a number of limitations on the position of the boundary are derived. With a boundary defined within these limitations, the experimental hydration energies cannot be reproduced, mainly because of the neglected specific interactions. In addition, we found that the description of the solute's electronic states also depends on the solvation model. We suggest that one or more explicit solvent layers are needed to obtain reliable solvation and excitation energies. © 1993 John Wiley & Sons, Inc.