Evaluations of soil organic carbon (SOC) stocks are often based on assigning a carbon density to each one of a number of ecosystems or soil classes considered, using data from soil profiles within these categories. A better approach, in which the use of classification methods by which extrapolation of SOC data to larger areas is avoided, can only be used if enough data are available at a sufficiently small scale. Over 190 000 SOC measurements (0-24 cm) have been made in the Flemish cropland (the Northern part of Belgium) in the 1989-2000 period. These SOC data were grouped into 3-year periods and as means plus standard deviation per (part of) community (polygons). This large dataset was used to calculate SOC stocks and their evolution with time, without data extrapolation. Using a detailed soil map, larger spatial groups of polygons were created based on soil texture and spatial location. Linear regression analysis showed that in the entire study area, SOC stocks had decreased or at best had remained stable. In total, a yearly decrease of 354 kton OC yr-1 was calculated, which corresponds with a net CO2 emission of 1238 kton CO2 yr-1 . Specific regions with a high carbon sequestration potential were identified, based on SOC losses during the 1989-2000 period and the mean 1999 SOC content, compared to the average SOC content of soils in Flanders with a similar soil texture. When restoring the SOC stocks to their 1990 level, we estimated the carbon sequestration potential of the Flemish cropland soils to be some 300 kton CO2 yr-1 at best, which corresponds to a 40-year restoration period. In conclusion, we can say that in regions where agricultural production is very intense, carbon sequestration in the cropland may make only a very modest contribution to a country's effort to reduce greenhouse gas emissions.