Forests play an important role in carbon sinks and mitigation of atmospheric concentrations of carbon dioxide and greenhouse effect. Given that sample plots used for collection of forest carbon observations are often much smaller than the map units of forest carbon at regional, national, and global scales, scientists are currently experiencing two challenges. The first challenge is to produce reliable maps of forest carbon using the data from inconsistent sizes of plots and image pixels. Also, because estimates of forest carbon normally contain uncertainties, the second challenge is to accurately model propagation of uncertainties from input data to output results. In this study, a methodology for mapping and analyzing spatial uncertainty of forest carbon estimates was developed to address these challenges. The methodological framework consisted of two methods. The first one was up-scaling method that combined and scaled up existing national forest inventory plot data and satellite images from smaller sample plots and image pixels to larger map units. The second one was spatial uncertainty analysis and error budget method that entailed modeling propagated uncertainties through a geostatistical mapping system. A case study using 46 permanent national forest inventory plots from Wu-Yuan County, Jiangxi, China, was undertaken to test this methodology. The results showed that this method reproduced not only the spatial distribution of forest carbon but also the spatial pattern of variances of its estimates and was able to quantify the contributions of uncertainties from the field plot data and satellite images to the uncertainties of forest carbon estimates. Thus, this study, to some extent, overcame the gaps that currently exist in the generation and assessment of forest carbon estimation maps. Moreover, the results showed that in this case study, the variation of the band ratio defined as (TM2 + TM3 + TM5)/TM7 contributed more uncertainties to the estimates of forest carbon than the variation of the plot data. In addition, we also found out that the product of the input plot forest carbon variance and the band ratio variance, implying the interaction between these two variables, reduced the uncertainties of the forest carbon estimates.