Hyperspectral images of soybean at different growth stages were collected by using multi-angular observation system in the field. Reflectance spectra of different objectives including soil, soybean plants, and illuminated leaves were extracted and they were used to retrieve the chlorophyll density of soybean canopy by using reflectance spectra at different observation angles, and furthermore explored detailed model inversion effects of soybean group for ground soil, shadow leaves, and angle changes. Studies have been showed that when the angles combination was 0°, 20°, 40° and 60°, there were the highest determination coefficient of 0.834, and the lowest root mean square error of 6.13 for prediction model of chlorophyll density. Meanwhile, the zenith angle and azimuth angle was 40° and 0°, respectively, which had the best inversion model. The changes of zenith angles were the most key factor to affect assessment accuracy of chlorophyll density for soybean canopy.