Digital image classification is a useful process to characterize the land cover and the land use by analyzing target’s spectral and spatial patterns. Digital image processing involves a manual or automatic process for the distinction of targets based on visual and statistical parameters. This study aimed to analyze the accuracy of two supervised classifications for OLI/Landsat 8 image featuring the island of Santa Catarina/Brazil applying two methods: the pixel-based approach by maximum likelihood classification with a post-processing step (majority filtering), and region-based approach by Bhattacharyya classification. Both classifications were submitted to an accuracy assessment through the validation of classes (urban, forest, pasture and bare soil, inland water and sand) based on references obtained from a high spatial resolution image (RapidEye). The highest agreement between classified image and the reference was achieved with the classification applying the region-based approach. This approach was then applied to evaluate the effects of the application of thermal band 10 (OLI/Landsat 8) in the classification process. Results revealed that the pixel-based approach and the region-based approach presented less accuracy in regions where the heterogeneity of the land cover is more complex and in addition, it was also revealed an agreement reduction between the classified data and the reference when OLI/Landsat 8 band 10 was used.