Abstract: Research on bidirectional reflectance characteristics of vegetation canopy is an important direction for quantitative remote sensing. The self-developed multi-angle observation system in this paper was used to collect imaging data of soybean in different sowing density from branch period to flowering period. Characteristics of changes in Bidirectional Reflectance (BR) for pure vegetation and mixed canopy, including vegetation and soil, were analyzed by hyperspectral images from segmenting vegetation, background soil, and shadow leaves. Studies have shown that observation position is the principal plane; canopy reflectance of pure vegetation after soil spectra is removed, gradually increased along with the decrease of zenith angle. When forward observation was conducted in principal plane, it differed from the canopy composed of vegetation and soil. When observation direction changed from backward to forward in the principal plane, canopy reflectance of pure vegetation in visible and near-infrared region showed a growing tendency gradually and the soil spectra was removed, this was different from before. In addition, when the observation position was in a perpendicular plane, there was a similar change for reflectance of soybean canopy before and after soil spectra was removed; but the symmetry of canopy reflectance of the former was better than the latter. The results had similar trends with BR change of soybean canopy in different sowing density. The paper provides a basis for the development of multi-angle remote sensing.