Monitoring vegetation condition by remote sensing is made difficult by the variation of canopy brightness (radiance) with view direction and sun position. In this paper, we develop a physical model of canopy radiance that models this variation as a product of three functions, each being expressed as a simple analytical formula. The first function, the S function, represents the proportion of canopy seen as sunlit, excluding the hotspot. The second function, the hotspot function, corrects the S function for the higher probability of seeing sunlit leaves near the anti-solar point. The third function represents the average radiance of sunlit leaves. The model was fitted to densely-sampled, multi-view radiance measurements of pine forest (Pinus radiata) and pasture (Lolium spp.), taken at a large (similar or equal to 60 degrees) sun zenith angle. Radiance predictions for multiple views were then made for a small sun zenith angle (similar or equal to 20 degrees). Comparison of predicted with measured radiances showed that the model out-performed the linear Roujean et al. model (J. Geo. Res., 97: 20,455-20,468) and the non-linear Verstraete et al. model (J. Geo. Res., 95: 11,755-11,765). The improvement is due to two factors: The hotspot function has a sun zenith angle dependence; and the S function allows the average projected leaf area to vary with off-nadir view angle. (C) 1997 Elsevier Science B.V.