The performance of density functional theory using the Perdew and Wang's exchange and correlation functionals (PW91) functional for the prediction of intermolecular interaction energies is evaluated based on calculations on the neon, argon, methane, ethylene, and benzene dimers, as well as on 12 hydrogen bonded complexes (water, methanol, formic acid, hydrogen fluoride, ammonia, formamide dimers and water-methanol, water-dimethyl ether, water-formaldehyde, hydrogen cyanide-hydrogen fluoride, water-ammonia, water-formamide complexes). The results were compared with those obtained from Becke's exchange and Lee, Yang, and Parr's correlation functionals (BLYP), Becke's 3 parameter functional combined with Lee, Yang, and Parr's correlation functional (B3LYP), second order Moller-Plesset perturbation (MP2), and coupled cluster calculations with single and double substitutions and with non-iterative triple corrections [CCSD(T)] calculations. The calculated interaction energies show that the PW91 functional performs much better than the BLYP or B3LYP functionals. The error in the computed binding energies of the hydrogen bonded complexes is 20% in the worst case. The most demanding cases are the systems with large dispersion contributions to the binding energy, such as the benzene dimer. In contrast to the BLYP and B3LYP functionals which fail to account for dispersion, the PW91 functional at least partly recovers the attraction. The basis set dependence of the PW91 functionals is relatively small in contrast to the MP2 and CCSD(T) methods. Despite its occasional difficulties with dispersion interaction, the PW91 functional may be a viable alternative to the ab initio methods, certainly in situations where large complexes are being studied. (C) 2001 American Institute of Physics.