A system for automatic detection of cephalometric landmarks is presented. Landmark detection is carried out in two steps: a line detection module searches for significant, well-contrasted lines of the image, such as the jaw line or the nasal spine. The landmark detection module uses the lines located in the first module to determine the search areas and then applies a pattern detection algorithm, based on mathematical morphology techniques. Relations between landmarks and lines are determined by means of a training process. The system has been tested for the detection of 17 landmarks on 20 images: more than 90% of the landmarks are accurately identified.