To identify the optimal spatial scales for which carbon agreements can be designed, the uncertainty in estimating stocks of soil organic carbon at field management scales must be understood. Thus, it was the objective of this study to compare alternative approaches for estimating the stocks of soil organic carbon of an agriculturally managed field. We sampled 17 ha of a uniformly managed field in Black Hawk Co., Iowa, where soils were developed in Wisconsinan loess. Soils occurred on characteristic landscape positions in a quintessential drainage sequence, ranging from well to poorly drained. The field was sampled in a regular grid pattern at 143 sampling locations to a depth of 30 cm. Total carbon and bulk density were determined for each of three depth increments. The stock of organic carbon was estimated in three ways: (1) by using a county-specific, expert database to predict soil characteristics, (2) by using classical statistical analyses, in which (a) the data from all sampling points are lumped or (b) in which the data from each mapping unit are stratified and then compared with those from other mapping units, and (3) by using geostatistical analysis, in which the data for the whole field are analyzed by semivariograms and a kriged map of carbon distribution is developed. Each approach resulted in a different estimate of the total organic carbon stock of the field. Each approach also yielded a different estimate of the uncertainty in the carbon stock value. Still, in this field of loess-derived soils, the predictions of organic carbon stock were relatively similar to one another (about 83 Mg organic C / ha).