The discrimination of different land uses based on satellite data is a widely used procedure, mainly because of processing speed and low cost of realization, depending on the sensor used. The process becomes complex when an accurate classification is desired, using many interrelated variables, referenced to a particular location, because is complex handling and understand such data. Thus, the study aimed to establish relationships between records of different spectral bands of the satellite Landsat-5/TM, considered independent variables, and the different land uses, regarded as cases, through multivariate technique called: multi-groups discriminant analysis. The radiometric calibration was performed, followed by the image rectification process. After this processing, it was obtained, utilizing the randomized method, the values of surface reflectance, for six spectral bands of the satellite, excluding the thermal band. The multi-groups discriminant analysis was utilized to classify the land uses. The identification of forests and water courses were adequately discriminated between them. There were difficulties to separate different types of uses related with pasture, due to water deficit, associated with the sensor spectral resolution.