Since its commercial introduction a decade ago, the technique of dual-energy X-ray absorptiometry (DXA) has been widely recognized as a useful and sensitive method of measuring changes in bone mineral density (BMD) at selected sites in the skeleton such as the spine and proximal femur. Because of their high precision and stable calibration, DXA scanners are frequently used in clinical trials to evaluate new treatments for osteoporosis. Quality assurance procedures based on regular scanning of phantoms are widely adopted in such trials, and continuity of the phantom BMD measurements is generally believed to ensure continuity in the in-vivo calibration. We report a change in calibration of a DXA scanner that occurred during a clinical trial where the calibration shift was different for the spine and femur sites and was not predicted or explained by the standard quality control procedures using phantoms. However, we show that provided patients enrolled in studies are thoroughly randomized and the statistical analysis is confined to the differences between the treated and control groups, then the effects of such calibration shifts on conclusions regarding the efficacy of treatment are considerably smaller than the random statistical errors.