We have compared, for atomic systems, the spherically averaged Fermi-hole correlation function p ! ( r l , s ) in the Hartree-Fock theory with the corresponding function p;i",r,, s ) employed in local density functional theory. It is shown that, in contrast to pQ,U(rl, s ) , the function pg", rl , s ) behaves qualitatively incorrectly at positions r l of the reference electron far from the nucleus. Furthermore, we have shown that the qualitatively incorrect behaviour of p;2E(rl, s ) can be remedied by an approximate expansion of p","(r I , s ) in powers of s , where s is the inter-electronic distance. However, such an expansion must be conducted in two regions due to the discontinuity of p",(rl, s ) as a function of s at the atomic nucleus. Based on the two-region expansion of p","(r,, s ) , we have developed an alternative approximate density functional expansionPQp&(rl, S) for the spherically averaged Fermi-hole correlation function. The corresponding exchange energy density functional yields values for the exchange energies of atoms in good agreement with Hartree-Fock results.Key words: atomic exchange energy, density functional theory, Fermi hole. I. Introduction Hohenberg and Kohn (1) demonstrated in 1964 that the ground-state energy of an electronic system is an unique (albeit unknown) functional of the electron density. This remarkable theorem gave credence to previous work by Slater (2) and others (3), and initiated an extensive study (4, 5) into the relation between electronic energy and electron density, a field that now has become known as (energy) density functional theory.In the last 15 years, due to a number of innovative implementations (6), it has become possible to carry out calculations based on approximate density functionals. Of the various approximate density functional methods, the HartreeFock-Slater (or X a ) scheme (7) is by far the most commonly used in chemistry. The accumulated experience with the HFS method seems to indicate that geometrical parameters (8), force constants (9), and spectroscopic energy differences (Sp, 10) (ionization potentials and excitation energies) are in good to fair agreement with experiment. However, bond energies (80-b, 1 l a ) are not in general well represented by the HFS method as indicated in Table 1, where several of the theoretical values obtained from HFS calculations typically are too high in comparison with experiment.The slightly more elaborate Local Spin Density method (LSD) (5p) provides geometrical parameters comparable to those obtained from ~l % calculations @a), whereas-ionization potentials and excitation energies from LSD calculations (12) compare somewhat better with experiment than the corresponding HFS results. However, the LSD method does not uniformly