Twenty-six profile silhouettes are divided by five judges in three categories: retrognathic, orthognathic, and prognathic. On the corresponding lateral cephalograms, 51 cephalometric variables are tested for profile recognition. Discriminant analysis indicates that measurements based on natural head position perform better than comparable variables based on an intracranial reference line. The soft tissue variables (SnPo'-SnTV, Po'-SnTV, ILS-SnTV) are most successful, but no single variable is totally accurate. Skeletal variables relating the maxilla to the mandible (APo, AB, NA-PoA) are more successful than those variables relating one arch to the remainder of the skull (A-FHP, B-NTV, SNA, SNB). A combination of soft tissue and dental variables (SnPo'-SnTV, SnPo'-TMe', 1u-Npo, 1l-NPo) is found to be completely successful. A methogological approach, as applied in this study, can pinpoint the cephalometric variables that are clinically more effective in describing certain facial features.