Reproducibility of lateral spine dual energy X-ray absorptiometry (LAT DEXA) scans using a Lunar DPX-L scanner was assessed in a cadaveric phantom and in patients. One hundred phantom measurements over 7 months demonstrated a longitudinal stability of 1.7% (coefficient of variation, CV). Additional scans were performed with the phantom rotated by up to 20 degrees in each of the three orthogonal planes to assess the effects of variable patient positioning. Horizontal and vertical rotation of the spine had little effect on the estimated bone mineral density (BMD), however, axial rotation of greater than 8 degrees led to errors in the BMD measurement. One hundred consecutive patients had two lateral scans performed within 1 month. BMD (range 0.10-1.6 g/cm2) was determined for each scan by one operator. Significant overlap from ribs and pelvis was often seen with L2 and L4 vertebrae but one vertebra (L3) could be measured in every case. Intraoperator and interoperator variability was assessed by three experienced operators, each analyzing 10 patients' scans on five separate occasions, and was found to be less than 1.1% for a single vertebra. BMD estimation of vertebral bodies and midslices by lateral DEXA scans (CV% of 3.8% and 4.6%) have a 95% confidence interval of 0.074 g/cm2 and 0.096 g/cm2, respectively for two vertebrae. This variability is due mainly to axial rotation, with operator variability, horizontal rotation, and vertical rotation having little effect on BMD estimation.