The reproducibility between observers of physical signs in patients with low-back pain was investigated. Fifty patients were examined by two surgeons and another sample of 33 patients was examined by a surgeon and a physiotherapist. Continuous data on five signs were analyzed by Pearsons' correlation coefficient, and binary data on 54 signs were analyzed by the Kappa agreement coefficient. Reliable signs consisted of measurements of lordosis and flexion range, determination of pain on flexion and lateral bend, nearly all measurements associated with the straight leg raising test, determination of pain location in the thigh and legs, and determination of sensory changes in the legs. Signs of root tension showed better agreement when qualified with a description of where the pain was experienced. Bony tenderness was more reliable than soft tissue tenderness.