Management zones (MZ) make economically viable the application of precision agriculture techniques by dividing the production areas according to the homogeneity of its productive characteristics. The divisions are conducted through empirical techniques or cluster analysis, and, in some cases, the MZ are difficult to be delimited due to isolated cells or patches within sub-regions. The objective of this study was to apply computational techniques that provide smoothing of MZ, so as to become viable operationally. Data physical soil properties and soybean yields were collected in 40 sampling points, being selected elevation and soil resistance to penetration (SRP) for the definition of MZ, because presenting spatial correlation with the normalized average yield. The software SDUM (Software for Definition of Management Zones) was used to perform descriptive statistical analysis, spatial correlation between attributes, data interpolation (inverse distance), and definition of MZ with two, three, four and five classes by clustering method Fuzzy C-Means. The MZ were generated with pixel size of 5x5 m, resulting in non-contiguous classes with patches or isolated cells, especially in maps with four or five classes. The reclassification process was carried out considering its close neighbors (3x3 and 5x5 pixels), making use mode and median statistics. It observed that the rectification of ZM through the implemented computational function was able to provide settings that facilitate the implementation of the AP and, in this study, the median statistic with 5x5 mask was the rectification method which demonstrated better results.