Technical Abstract: The landscape of the northeastern United States is diverse and patchy, a complex mixture of forest, agriculture, and developed lands. Many urgent social and environmental issues require spatially-referenced information on land use, a need filled by the National Land-Cover Data (NLCD). The accuracy of the NLCD data from 1992 and 2001 was assessed by aggregating it by county and comparing proportions of cropland and pasture/hay area to those recorded in the USDA Census of Agriculture (CoA) for the corresponding years. Estimates of total agricultural area corresponded closely between the two data sources (r2 = 0.91 for 1992 and 0.85 for 2001/2), but NLCD performed poorly at distinguishing between cropland and agricultural grassland (r2 = 0.64 and 0.80 for cropland, and only 0.50 and 0.44 for agricultural grassland). High forest cover reduced accuracy. Landscape complexity and patch size may also impede classification accuracy. Accuracy varied greatly by state, even for states within the same mapping regions. The NLCD is the best-available source of spatial information on agricultural land uses, but may not be suitable for projects requiring that agricultural lands be divided into subclasses.