Abstract Previous studies on model Lennard-Jones systems have shown that conformal solution theories tend to fail when mixtures contain molecules with large size differences. This failure can be attributed to deficiencies in the underlying mean density approximation (MDA) for the mixture's radial distribution functions which was originally proposed by Leland and coworkers. In this communication, we present an improved mean density approximation which is obtained from a straightforward application of statistical mechanical perturbation theory. Comparisons of the new theory with simulation PVT data for systems with large size ratios show it to be far superior to van der Waals one-fluid and MDA n-fluid theories, and an improvement upon hard sphere expansion and perturbation theories as well. Comparisons are also made with chemical potential simulation data, and for these data the new theory is competitive with perturbation theory.