Hyperspectral remote sensing data offer unique opportunities for the characterization of the land surface and atmosphere in the spectral domain. However, few studies have been conducted to estimate albedo from such hyperspectral data. In this study, we propose a novel approach to estimate surface shortwave albedo from data provided by the Airborne Visible Infrared Imaging Spectrometer (AVIRIS). Our proposed method is based on the empirical relationship between apparent directional reflectance and surface shortwave broadband albedo established by extensive radiative transfer simulations. We considered the use of two algorithms to reduce data redundancy in the establishment of the empirical relationship including stepwise regression and principle component analysis (PCA). Results showed that these two algorithms were able to produce albedos with similar accuracies. Analysis was carried out to evaluate the effects of surface anisotropy on the direct estimation of broadband albedo. We found that the Lambertian assumption we made in this study did not lead to significant errors in the estimation of broadband albedo from simulated AVIRIS data over snow-free surfaces. Cloud detection was carried out on the AVIRIS images using a Gaussian distribution matching method. Preliminary evaluation of the proposed method was made using AmeriFlux ground measurements and Landsat data, showing that our albedo estimation can satisfy the accuracy requirements for climate and agricultural studies, with respective root-mean-square-errors of 0.027 compared with AmeriFlux and 0.032 compared with Landsat. Further efforts will focus on the extension and refinement of our algorithm for application to satellite hyperspectral data.