To identify the changes of bony landmarks with rotation and to compare the reliability and accuracy of Perdriolle's method before and after training on the exact bony landmarks.

Simple A-P radiographs were taken with rotation from 5 degree to 50 degree at 5 degree interval using 10 dry human vertebrae (2 sets of T3, T6, T9 8 L3) with no bony abnormality. Axial rotation was measured before and after training on the exact bony landmarks by three observers, twice for each radiographs. Repeated measures of ANOVA were used for inter-< intraobserver variance and regression analysis for root mean square error and deviation of differences.

Mamillary process, inner margin of lamina, proximal and distal costovertebral joints, upper margin of concave pedicle and tip of the transverse process on the convex side confused the exact outline of the pedicle, while neural foraminal margin was helpful in identifying the location of the indistinct convex pedicle and concave vertebral border. Of the 660 measurements before and after training 372 and 397 errors were made, respectively, of which 88.9% and 88.5% were within ±5 degree(P > 0.05). Incidence of error greater than 5 degree was high at upper thoracic vertebra and 20 degree-40 degree of rotation regardless of training. The effects of training, however, was greatest at T3 and 20 degree-40 degree of rotation. Inter-< intraobserver variance was diminished at T3 and at the rotation between 20 degree-40 degree after training.

We could conclude that training on the exact bony landmarks did not reduce the overall error incidence. Accuracy and reliability, however, was increased at T3 after training and the efficacy of training was more evident in moderate vertebral rotation than in small or large vertebral rotation.