As part of a longitudinal comparison of bone mineral density (BMD) results originally obtained using a Lunar dual photon absorptiometry (DPA) scanner and later, using a Lunar dual X-ray absorptiometry (DXA) scanner, we compared femur results between DPA and DXA according to DXA analytic software (versions 1.3y and 1.4), and according to the method of placement of the femoral neck box (software algorithm or operator placement according to the appearance of the pair of images) in 58 elderly men and women. The mean BMD at each of three femoral sites was higher using DXA version 1.3y than DPA, but the use of software version 1.4 brought the BMD value closer to that of DPA at all sites. Of 58 scans, 12 (21%) were changed by the operator, resulting in an overall reduction in mean percent BMD difference between scan pairs of 79% (from 1.24% to 0.29%). Although the differences between the DPA/DXA software-driven analysis and the DPA/DXA operator-driven analysis appeared small (high r2 values and intra-class correlation coefficients), the increase in sample size that would be required for the same power to detect 2-year changes in BMD if the software-driven analysis was used instead of taking the time to perform the operator-driven analysis was 18%. The findings of this study highlight the need to account for upgrades in analytic software. Furthermore, we present a rational approach for the analysis of serial scans that has face validity and that results in smaller differences between pairs of scans performed on the same individual. The decision to adapt these methods must be based on the relative costs of reducing unwanted scan variability.