Soil is an essential part of any terrestrial ecosystem. Scientists, technicians, and farmers have studied its physical and chemical properties for many years for agriculture and soil conservation. These studies usually require field sampling and laboratory analysis that are time-consuming and destructive to the samples being analyzed. Remotely sensed data are an alternative that provide reliable information at low cost based on a non-destructive technique. The objective of this study was  to evaluate the usefulness of Landsat Thematic Mapper data to classify soils in arid lands. To this end, a Thematic Mapper (TM) scene from the Chihuahuan Desert at Doña Ana County, New Mexico, mapped with the Soil  Taxonomy System, was used. Furthermore, four remote sensing approaches were created to determine the best method to identify soil-mapping units. They were named simple, technical, scaled, and complex. The agreement of TM data and soils maps was tested using the error matrix approach in a supervised classification. Spectral signatures were selected by separability analysis applying the transformed divergency technique. The results revealed that the simple approach, based on thermal band discrimination, obtained classification accuracies of 70.67%, suggesting bands 2, 4, and 7 as the best for identifying soil mapping units. The technical approach, based on the principal components analysis technique, obtained accuracies of  66.86%, suggesting that data reduction is possible through this technique.