The heterogeneity in surface cover caused by management practices or natural events complicates monitoring the conditions of grasslands and asseessing their productivity by remote sensing techniques. Statistical procedures were sought that would allow different grassland surface cover types (bare soil,senescent vegetation, and green vegetation) to be distinguished by using measurements of grassland multispectral reflectance. Two procedures, discriminant functions and canonical discriminant analysis, were found suitable for achieving this objective. Linear classification functions and canonical variables were derived, which distinguish between the three cover types. A comparison between two sensor systems, a Barnes multiband radiometer and an Exotech radiometer that simulates the Landsatthematic mapper (TM) and multispectral scanner (MSS) bands, respectively, showed that the separability among the three cover types was substantially improved by the additional and improved spectral features of the Barnes radiometer