Agricultural activities have dramatically altered our planet’s land cover. To understand the extent and spatial distribution of these changes, we have developed a new global data set of croplands and pastures ca. 2000 by combining national and sub-national agricultural inventory data and satellite–derived land cover data. The agricultural inventory data, with much greater spatial detail than previously available, is used to train a land cover classification data set obtained by merging two different satellite-derived products. By utilizing the agreement and disagreement between Boston University’s MODIS global land cover product and the GLC2000 data set, we are able to predict the spatial pattern of agricultural land better than by using either data set alone. We present a new global 5 min (~10 km) resolution cropland and pasture dataset for the Year 2000 that is of greater accuracy than previously available, and for the first time, statistical confidence intervals on these estimates.