Green's concept of an equivalent-source layer is invoked to construct a data adaptive, zero-phase, Wiener band-pass filter for regional-residual gravity anomaly separation. The observed gravity field's Fourier power spectrum is modeled with two Green's equivalent-source layers, one equivalent layer for the shallower (residual field) geologic sources, and a second equivalent layer for the deeper (regional field) geologic sources. The depths and average physical property contrasts of the two equivalent layers are determined by fitting the observed gravity field's Fourier power spectrum with a two-layer spectral model. Each equivalent layer is simulated by a horizontal thin sheet with randomly varying and randomly distributed point density sources spread throughout it. Adopting such a theoretical model for the Fourier power spectrum yields a stable and well-behaved filter transfer function. Like all band-pass filtering though, the method is ineffective in the case of insufficient vertical separation between the shallow and deep geologic sources whose gravity anomalies it is desired to separate. The filter design process is simple and easily automated, being well-suited for modular implementation in a 'filter tool kit' applications program running on either a workstation or personal computer. Equally important, the method yields repeatable, interpreter-independent results.