A new method for the measurement of scoliotic curves in antero-posterior (AP) radiographs is presented, in which the centre of the surface image of the vertebral bodies of the apical and two end vertebrae of the curvature are defined on the basis of geometric principles. Measurements using the Cobb, the Ferguson, and the new method were performed on ten AP radiographs from each of three groups of young patients with right convex thoracic idiopathic scoliosis with Cobb angles of between 7 and 15 degrees, 16 and 45 degrees and 46 and 80 degrees, respectively. Measurements using the Cobb method yielded significantly higher values than measurements using either the Ferguson method or the new method. In curves with Cobb angles of between 7 and 15 degrees, the values using Ferguson's method were significantly lower than those using the new method; the difference increased significantly in curves with a Cobb angle of 16 degrees or more. The level of significance of the intra- and interobserver differences between the new, the Cobb and the Ferguson methods was significantly higher in curves with a Cobb angle of 16 degrees or more. It is argued that measures of the scoliotic angle obtained by the new method are of greater clinical relevance than those obtained by the two other methods. Unlike the Cobb method, the new method takes into consideration the translation of the apical vertebra in relation to the end vertebrae and not only the tilt of the end vertebrae of the curve. As compared to the Ferguson method, the new method is based on standardised geometric principles, and is not influenced by changes in the shape of the vertebral body. Moreover, the repeatability of the new method is greater than that of both the Cobb method and the Ferguson method. Therefore, it is believed that the new method provides a more accurate measure of the scoliotic curve than do the two other methods, and it is to be preferred over the other two methods in longitudinal evaluation of the development of the curve.