The bending properties of cadaveric lumbar spines were measured and used to convert in vivo measurements of lumbar flexion into bending moments ('stresses'). Forty-two lumbar motion segments were subjected to complex physiological loading and graphs were obtained of bending moment vs flexion angle. Variability was reduced by expressing both variables as a percentage of their values at the elastic limit. Data were averaged for each lumbar level, and a composite bending curve was compiled for the lumbar spine, L1-S1. A linear relationship was established between lumbar flexion measured in vitro and in vivo. This enabled values of 'per cent lumbar flexion' measured in vivo to be converted into 'per cent maximum bending moment' with a maximum likely error of about +/- 8%, which is equivalent to about +/- 5 Nm at L5-S1 for an average person. The technique was applied to 28 subjects, using dynamic measurements of lumbar flexion obtained with the '3-Space Isotrack' system. The bending moment at L5-S1 was 12 Nm on average when picking a pen up off the floor. Highly significant increases in bending moment were observed when heavier and bulkier objects were lifted.