Physically-based retrieval of vegetation canopy properties from remote sensing data presumes a knowledge of the spectral albedo of the basic scattering unit, i.e. leaf. In this paper, we present a novel method to directly retrieve the spectral dependence of leaf single-scattering albedo of a closed broadleaf forest canopy from multiangular hyperspectral satellite imagery. The new algorithm is based on separating the reflected signal into a linear (first-order) and non-linear (diffuse) reflectance component. A limitation of the proposed algorithm is that the leaf single-scattering albedo ω(λ) is retrieved with an accuracy of a structural parameter (called a0) which, in turn, depends on canopy bidirectional gap probability, ratio of leaf reflectance to transmittance, and distribution of leaf normals. The structural parameter (a0) was found to depend on tree-level structural parameters, such as tree height and volume of a single crown, but not the amount of leaf area.