This paper provides an overview of multiband classification tools and their application to Earth remote sensing, medicine, and environmental geophysics. The methods discussed are based on the assumption that natural objects can be uniquely associated with the variation of their electromagnetic properties along a wavelength. Similar objects are identified with spectral classes defined as clusters in a feature space, and exploratory methods provide ways of defining these clusters without a priori knowledge of the data cloud structure. Spectral classes are statistically characterized from a representative sample of objects surveyed on the ground, and this knowledge is then applied to the whole image in order to build a final classification map.