Net solar radiation (Q⋆) and the partitioning of Q⋆ between vegetation (Qveg⋆) and the substrate (Qsoil⋆) are important quantities computed by soil-vegetation-atmosphere-transfer (SVAT) models. Commonly, two-stream models of canopy radiative transfer are used for this purpose. We examine the validity of this approach for nonuniform canopies and compare estimates of Q⋆, Qveg⋆, and Qsoil⋆ computed using a two-stream model with estimates computed using a model that accounts for the effects of three-dimensional (3-D) structure in vegetation on radiative transfer. To accomplish these goals, a sensitivity analysis is conducted for the key input parameters to two-stream canopy radiative transfer models. The parameters examined include leaf area index (L), leaf optical properties, solar zenith angle, leaf orientation, and background albedo. Sensitivity analyses are also conducted using a Geometric-Optical Radiative Transfer (GORT) model. The GORT model treats vegetation canopies as being composed of 3-D crowns and allows for both vertical variation in leaf area and horizontal variation in stem density. Results show that Q⋆ computed by the two-stream model varies by up to 10% relative to the GORT model. Further, the partitioning of Q⋆ between the vegetation canopy and substrate computed by the two-stream model can vary by up to 30% relative to the more realistic GORT model, even for relatively dense canopies (L=6). These differences arise because two-stream models use parameterizations for gap probabilities that are not realistic in discontinuous canopies. Based on this conclusion, a parameterization is proposed to include the effects of canopy heterogeneity in two-stream models. Results from the original two-stream model, a parameterized two-stream model, and the GORT model are compared using data from the Boreal Ecosystem-Atmosphere Study. These results show that the proposed parameterization captures the effect of 3-D structure in vegetation on radiation regimes and is therefore suitable for inclusion in SVAT models. Inclusion of improved treatment for radiative fluxes in land surface parameterizations should improve modeled estimates for other flux quantities computed by SVAT models. ¿ 2001 American Geophysical Union