The reliability of predicting the results of orthodontic surgical treatment was analysed. The sample consisted of 30 patients treated with one of the following six types of surgery: mandibular anterior alveolar surgery; maxillary anterior alveolar surgery; mandibular setback surgery; mandibular advancement surgery; Le Fort I surgery; and a combination of Le Fort I and mandibular setback surgery. Comparisons of tracings of the initial cephalometric radiographs, the prediction tracing and the six-month follow-up film showed great variation in the results both within and between the surgical subgroups. Generally, it was easier to predict the treatment results of alveolar segmental osteotomies, especially in the mandible, than of operations in which the whole mandible was repositioned. The results from Le Fort I surgery with or without a concomitant mandibular setback showed the greatest difference between the predicted and the actual outcome. The postoperative vertical dimension appeared to be particularly hard to predict. Explanations for these discrepancies are offered and possibilities for improvements suggested. It is concluded that prediction tracings are still of value despite the poor accuracy in some cases.