We present a new paradigm for the design of exchange-correlation functionals in density-functional theory. Electron pairs are correlated explicitly by means of the recently developed second order Bethe-Goldstone equation (BGE2) approach. Here we propose a screened BGE2 (sBGE2) variant that efficiently regulates the coupling of a given electron pair. sBGE2 correctly dissociates H2 and H + 2 , a problem that has been regarded as a great challenge in density-functional theory for a long time. The sBGE2 functional is then taken as a building block for an orbital-dependent functional, termed ZRPS, which is a natural extension of the PBE0 hybrid functional. While worsening the good performance of sBGE2 in H2 and H + 2 , ZRPS yields a remarkable and consistent improvement over other density functionals across various chemical environments from weak to strong correlation.The popularity of density-functional theory in physics, chemistry and materials science stems from the favorable balance between accuracy and computational efficiency offered by semi-local or hybrid approximations to the exchange-correlation (xc) functional. However, certain well-documented failures such as the unsatisfactory prediction of atomization energies, the significant underestimation of weak interactions and reaction barriers and the inability to correctly describe strongly interacting scenarios with pronounced multi-reference character, such as bond dissociation [1-7], limit the predictive power of these functionals in certain cases.Density functionals that depend on the unoccupied as well as the occupied Kohn-Sham orbitals stand on the fifth and currently highest rung of the ladder [8] of density functional approximations. The rapid growth of computational capacity has been boosting the development of practical level-5 functionals over the past ten years. One example is Görling-Levy perturbation theory at 2nd order that corresponds to the exact xc-functional for systems with a linear adiabatic-connection path [4,9]. However, in reality the adiabatic-connection path is not linear and Görling-Levy perturbation theory fails for systems with small energy gaps, where (near)-degeneracy correlation (also known as static correlation) is dominant, as exemplified by molecular dissociation [5][6][7]. The random-phase approximation (RPA) is another example of a level-5 functional. RPA sums up a sequence of "ring diagrams" to infinite order [10] and is remarkably accurate for reaction-barrier heights and weak interactions, but it significantly underestimates atomization energies. It also produces the correct H 2 dissociation limit [11], but fails for H + 2 dissociation due to appreciable self-correlation errors [12,13]. Recently, much effort