The interobserver reliability of lumbar spinal signs is low for many signs but little has been published on the ability of training to improve this. This small study describes the interobserver reliability of some common lumbar spinal signs and the effect of a brief training intervention on this reliability. Four subjects with low back pain underwent a lumbar spinal examination by four doctors. Posture, pain on movement, leg length inequality and tenderness was tested. After a 40 minute session comparing findings an d examination techniques, the four doctors examined a different four subjects with low back pain. Kappa coefficients of agreement on the eight subjects' findings were moderate for scoliosis, poor for iliac crest height and only fair for the forwards flexion test. Agreement was substantial to almost perfect for pain ion movement tests. Agreement for tests of leg length inequality was slight to moderate and was moderate to substantial for regional tenderness. Agreement was essentially unchanged before and after training for fourteen of the eighteen tests. However, agreement decreased after training for forward flexion test, pain on flexion, left iliac crest tenderness and left buttock tenderness. These decreases may have resulted from small sample sizes and differences in the subjects between sessions. Nonetheless, the results suggest that far more than 40 minutes training is required to improve agreement between doctors examining the lumbar spine.