Dual-energy X-ray absorptiometry (DXA) measurements were analyzed using two versions of software (Hologic V8.1a and V8.21) to compare the short- and long-term precisions of the measurements. Software V8.21 was designed by the manufacturer to better address magnification effects on estimations of soft tissue lean mass. Twenty weight-stable, obese postmenopausal Caucasian women aged 40-70 yr participated in the study. Total and regional body composition measurements were obtained at baseline and after 3 mo, using a fan beam Hologic QDR 4500A absorptiometer. For the estimation of precision, duplicate scans obtained on the same day for nine women were analyzed using both versions of the software. The correlations between duplicate scans ranged from 0.886 to 0.998 and were similar between software versions. The CVs for fat and lean weights and bone mineral content (BMC) were 1.2%, 1.1%, and 1.7%, respectively, for software V8.21 compared to 1.3%, 1.3%, and 2.1%, respectively, for V8.1a. Systematic differences were found between software versions with higher values for fat and lean weights for software version V8.21. The 3-mo, long-term reproducibility of body composition estimates from DXA was only slightly less than short-term reproducibility for both software versions (coefficient of variation [CV] range from 1.3% for BMC weight to 11.0% for arm fat). Software V8.21 yielded smaller percentage mean differences between scale and DXA-estimated weights (-2.4% and -7.2% at baseline and -2.9% and -7.6% at 3 mo, respectively) and higher fat and lean weights (49.12 and 47.1 kg and 49.6 and 44.6 kg, respectively) than V8.1a. Reproducibility of all variables was comparable between software versions.