Absolute values of bone mineral density (BMD), using dual-energy X-ray absorptiometry (DXA), differ between instruments from different manufacturers. Despite these differences, the rates of change calculated from serial measurements on different densitometers have been assumed to be comparable. We compared the change in BMD in 34 subjects in the lumbar spine and at the standard sites in the proximal femur, from measurements performed using a Lunar DPX-L and a Hologic QDR-1000. Measurements were obtained on the same day, and repeated on the same day, on both machines after a mean interval of 4.8 years (range 4.1-6.3 years). There were strong positive correlations between the percentage change in BMD calculated using the two machines in the lumbar spine, trochanteric region, and total proximal femur: r = 0.82, 0.84, and 0.73, respectively (p < 0.0001 at all sites). In the femoral neck and in Ward's triangle, the correlations were not as high: r = 0.55 (p = 0.003) and 0.43 (p = 0.028), respectively. At all sites, despite the significant correlations, the agreement between the two densitometers was not high and there may be significant errors in individual subjects if one uses measurements from one densitometer to predict the change in BMD using the scanner of the other manufacturer. There is less of a problem comparing group data in the lumbar spine and trochanteric region, although errors are still likely to occur in comparing group data of bone loss, calculated using different densitometers, in the other proximal femur sites. In conclusion, the study suggests that caution is necessary in combining bone loss data derived using densitometers from different manufacturers, particularly in the proximal femur. This has important implications for multicenter studies.