To address the problem of unmixing spectral mixtures of vegetation and soils in remote sensing data a numerical model was developed to quantify the NIR light scattering between green vegetation and a background. When tested using laboratory measured multispectral images the numerical model was found to be accurate. According to the model, NIR sidescattering by green leaves contributed a significant amount of light to adjacent, unshaded background while NIR light transmitted through the leaves increased model leaf reflectance. Furthermore, NIR transmittance/scattering produced non-linear spectral mixtures in which the non-linearity increased with an increase in leaf transmittance or background reflectance. A linear model, when applied to simulated mixtures using spectrally flat soils, overestimated the green leaf fractions, underestimated the shade fractions and accurately estimated the background fractions as long as shade was included as an endmember. Linear estimates from simulated mixtures of real soils produced a moderate overestimate ($<$8%) of the soil fractions. The analysis was extended to Airborne Visible and Infrared Imaging Spectrometer (AVIRIS) data collected over Jasper Ridge, California. Most of the spectral variability in the scene was explained by mixtures of three endmembers, green vegetation, soil and shade. Soil types and non-green vegetation were distinguished using residual spectra produced by absorptions that were absent in the endmembers. Green vegetation types were distinguished using a linear model applied separately to wavelength subsets of the entire spectrum. As predicted by the numerical model, NIR scattering significantly affected abundance estimates of green leaves and shade in the image. A new technique was developed for improving abundance estimates of green vegetation and shade by solving for canopy shade. The shade fractions and canopy shade spectra varied with vegetation type. Coupled with the numerical simulations, the image analysis provided new methods for qualitatively identifying soils and vegetation and providing accurate estimates of abundance.