Test particles infalling from infinity onto a compact spherical star with a mildly super-Eddington luminosity at its surface are typically trapped on the "Eddington capture sphere" and do not reach the surface of the star. The presence of a sphere on which radiation pressure balances gravity for static particles was first discovered some twenty five years ago. Subsequently, it was shown to be a capture sphere for particles in radial motion, and more recently also for particles in non-radial motion, in which the Poynting-Robertson radiation drag efficiently removes the orbital angular momentum of the particles, reducing it to zero. Here we develop this idea further, showing that "levitation" on the Eddington sphere (above the stellar surface) is a stateof stable equilibrium, and discuss its implications for Hoyle-Lyttleton accretion onto a luminous star. When the Eddington sphere is present, the cross-section of a compact star for actual accretion is typically less than the geometrical