The use of Synthetic Aperture Radar (SAR) data, sole or fused/combined to optical data, is interesting in regions like the Brazilian Amazon, in which cloud cover is constant. Many works has been done in the region, comparing classifications obtained from SAR, optical or fused data. In general, optical data has shown better results than SAR data. Fused data has either presented better, comparable or worse results. This paper presents land cover classifications for a region in low Tapajós River, in the Brazilian Amazon, using optical, SAR and fused data, using three legends, with different level of detail (number of classes). It was observed that the lower is the number of classes, the higher is the accuracy of the classification. Although optical results were better for all legends level, results obtained from SAR data is comparable to those when using a small number of classes. In addition, some classes where better classified using specific type of data.