In the assessment of vertebral bone mineral density (BMD) with quantitative computed tomography, accuracy is limited by systematic errors caused by beam-hardening effects and unknown marrow and fat concentrations in trabecular bone. Reproducibility depends on precise repositioning of the patient, selection of the region of interest (ROI), and apparatus stability. In an integrated approach, these error sources were taken into account. Dual-energy methods are incorporated and tested to avoid systematic errors. Contour-detection algorithms for automated ROI determination that work reliably and offer improved reproducibility and efficiency are presented. A new design of reference phantom with solid tissue-equivalent materials provides calibration in BMD units and checks apparatus stability. Dual-energy methods are necessary to achieve accuracy of 5% or better. Automated ROI definition and reference phantom methods are mandatory for reproducibility of 2% or better.