Bone mineral density (BMD) measurement is a major outcome measure in osteoporosis. The BMD changes observed must exceed the variability inherent in the measurement process to be considered related to disease progression. The objective of the study was to estimate short-term variability of BMD measurement and to propose a cut-off value for the smallest detectable BMD changes for an individual. To estimate the short-term variability, 70 healthy postmenopausal women aged 53 +/- 4 years (group 1) and 57 elderly osteoporotic postmenopausal women aged 80 +/- 6 years (group 2) had two repeated BMD measurements of the lumbar spine (L2-L4) and the proximal femur with dual-energy X-ray absorptiometry, with complete repositioning within 1 h. Cut-offs derived from short-term variability were either estimated from the coefficient of variation (CV) (which is a function of the measured value) or from the standard deviation (SD), and applied to 330 postmenopausal women (group 3) who had BMD measurements at baseline and 2 years later. The short-term intrasubject variability was greater at the lumbar spine in group 2 versus group 1 (0.0123 vs. 0.0059 g/cm2, p < 10-4), whereas it was not at the femoral neck (0.0098 vs. 0.0076 g/cm2, p = 0.28). There was no statistically significant correlation between short-term intrasubject variability (SD) and BMD as demonstrated with an analysis of covariance (p values ranging from 0.17 to 0.90). Cut-offs estimated with SD and CV were individually applied to group 3 patients. Using these two cut-offs, discrepancies in assessment of progression were observed in 1.7-8.6% of cases. Short-term BMD variability is constant in a wide range of BMD values. Consequently, to determine cut-off values for the smallest detectable differences in BMD at the individual level, precision errors should be based on SD (expressed in absolute units) rather than on CV (expressed in percentage).