We examined sources of error in bone measurements made with computed tomography (CT) using a whole-body scanner( GE 8800) and a peripheral-bone CT scanner (developed at the University of Alberta). We investigated the influence of various factors on trabecular bone density: homogeneity and noise in the image plane, linearity of calibration, body size, effects of cortical bone, and the image analysis procedure. With the GE 8800 scanner, the precision (SD) of measurements of a single vertebra is expected to be ±1.65% (noise: ±0.22%, calibration: ±1.3%, analysis: ±1%); the accuracy, excluding consideration of marrow fat, varied between -2.7 and +7.3% (compact-bone thickness: 2-5%, body size: -2.5-+1.5%, calibration: -0.47-+0.77%). With the peripheral-bone CT scanner, the total precision error (±0.53%) was dominated by noise, with only a minor contribution from the analysis procedure (±0.04%); accurary varied between -0.6 and +3.4% (effect of cortical bone: up to 3.0%; changes in size of object: -0.59-+0.4%). The magnitude of these errors was determined under ‘ideal’ conditions, mostly through phantom measurements; therefore, the errors represent optimistic lower limits in clinical application. Furthermore, measurements of density of cortical bone were not reliable for bone thicknesses of less than about 4 mm with the GE 8800 scanner and less than about 1.5 mm with the peripheral scanner.