An unsupervised method is proposed for detecting circular structures on satellite images partly based on a version of the Hough Transform. This method consists of two major steps: the extraction of significant contours from the image, using mathematical morphology operators, and the detection of circular structures from the contours previously denned. The latter includes four substeps: individualization of the curves, decomposition of the curves into near-circular elements, application of the ‘adaptive HT’ version to each near-circular element, and computation of more precise results. As an example, the method is applied to a SPOT scene in Southern Peru to detect geological structures.