Observations in remote sensing generally result in compositional measurements due to the finite sampling aperture of the sensor. The measurements are modeled as mixtures of distinct endmember signatures. Unmixing algorithms attempt to estimate the mixture components and their proportional contributions to the measurement, and by inference, to the composition of the measured scene. The choice of an unmixing algorithm depends on our knowledge about the mixture process, the number of endmembers present and their unmixed signatures, the variational structure of the signatures and proportions, and sources of nonsystematic errors and noise. A number of linear unmixing algortihms are surveyed and tested against synthetic hyperspectral reflectance data under a variety of conditions reflecting varying degrees of uncertainty in the endmember population and system noise structure.