In this paper we present a novel method of considering self-occlusion when fitting faces under global illumination. The effects of self-occlusion under spherical harmonics (SH) illumination are predicted through the use of a a morphable model of faces. A linear function is precomputed which maps morphable model coordinates directly to surface irradiance values. We demonstrate that considering self-occlusion yields a clear improvement over an otherwise identical reconstruction where self-occlusion is neglected. Our method succeeds at reconstructing facial attributes that are visible primarily as a consequence of self-occlusion, such as the depth of the eye sockets and the shape of the nose.