Twenty-two measurements repeated non-sequentially on each of 10 patients by five observers were undertaken to determine their reliability for routine clinical use. Measurements without significant inter-observer variation or with a coefficient of reliability greater than 0.70 were cervical rotation, cervical lateral flexion, tragus to wall distance, fingertip to floor distance on sagittal and lateral flexion, C7 to iliac crest line distraction and modified Schober index. It is concluded that many of the currently used measurements are either statistically unreliable or clinically unhelpful in mild or moderate ankylosing spondylitis. The most clinically useful were cervical rotation using a protractor, cervical lateral flexion using a goniometer, thoracolumbar flexion as the C7 to iliac crest line distraction, thoracolumbar lateral flexion as the fingertip to floor distance and the modified Schober index.