In this study digital images of bone cross-sections obtained by computed tomography were analyzed with an automated outlining method. It was shown that unbiased cross-sectional geometric measurements of cortical bone could be obtained if the periosteal and endosteal surfaces were defined at separate thresholds. Use of different threshold levels for these two surfaces resulted in errors of 2.6% for periosteal diameters, 7.4% for endosteal diameters and 7.3% for cortical area. If incorrect thresholds were used, cortical thickness measurements can have errors as high as 30%. In addition, simulated variation in medullary fat content did not affect measurement of medullary dimensions.