High-resolution, remotely sensed imagery with bare soil surface has been successfully used to quantitatively map the spatial variation of the organic carbon concentrations (SOC) of surface soil (Chen et al. 2000, Chen et al 2005). This method requires each field to be sampled and mapped separately. Comparing with the traditional method in precision agriculture such as gird sampling, the method greatly reduced the cost of soil sampling and the following data analysis. However, the cost would be still high if multiple fields need to be mapped. Upon examination of remotely sensed images that consist of several fields, some fields look similar with regard to image properties such as the color histogram. By examining the similarity of these fields, it may be possible to group the similar fields, analyzing and mapping them together in a single procedure, thereby reducing costs further. The objective of this study was to examine image similarity of agricultural fields, relate image similarity to soil properties, and map SOC for a group of fields with a single procedure. Three types of features (feature vectors), including color histograms, color slope magnitudes, and Haar wavelet transforms, were extracted to analyze field similarity with computer programming developed. Two similarity matching methods, statistical clustering with Euclidean distance and the Ward neural network system were used for examining the similarity between extracted features extracted from fields. Dissimilarity distance (for the statistical clustering method) and coefficient of determination (for the Ward neural network system) were used as criteria of similarity ranking of images (fields). Based on the conclusion of similarity analysis, different fields were selected from an NASA ATLAS image. Similarity of fields was examined and soil organic carbon maps for grouped fields were developed with a single processing for each group of fields.