It is well known that C Band SAR data is not adequate for land cover classification especially in tropical environments. The purpose of this paper is to study selected feature extraction techniques to improve the C band usefulness. ERS1/2 tandem mode data was acquired over the Tapajos National Forest, Brazil, and gave the opportunity of testing the C band coherence map for classification purposes. The use of coherence for land cover classification is justified since it is expected low coherence in forest areas, in comparison with bare soil and sparse vegetation, which have high coherence. Texture is also a standard feature normally employed and from a set of fourteen of the Haralick's features, one was selected to be tested jointly with C band coherence map and the C band backscatter itself. Two sets of classes, one with 10 classes, the other with 4 classes were defined for training the classifiers. These classes represent typical classes in the region, and mainly include primary forest, several stages of regeneration and pasture. Iterative Conditional Mode (ICM), which is a contextual supervised per point classification scheme, and a supervised region classifier from a segmentation map produced by region growing type of segmentation were used to produce the classification maps. Statistical tests based on Kappa statistics were used to test the precision and significance of the results. The result shows that coherence map alone is adequate for the case of four classes, and the inclusion of any other feature either does not change the precision significantly or worse the results.