In this paper, we present a novel, hardwareaccelerated approach to compute the visibility between surface points and directional light sources. Thus, our method provides a first-order approximation of the rendering equation in graphics hardware. This is done by accumulating depth tests of vertex fragments as seen from a number of light directions. Our method does not need any preprocessing of the scene elements and introduces no memory overhead. Besides of the handling of large polygonal models, it is suitable for deformable or animated objects under time-varying high-dynamic range illumination at interactive frame rates.