The reproducibility of dual-energy X-ray absorptiometry (DXA) measurements of bone mineral density (BMD) is an important factor for longitudinal studies. We assessed the long-term precision of posteroanterior lumbar spine, femoral neck and total hip BMD in 40 postmenopausal women who formed the control arm of a clinical trial of tibolone. BMD was measured at 0, 6 and 12 months and thereafter every 12 months up to 7 years. For each subject the trend of BMD with time was analyzed using linear regression. Each residual was expressed as the percentage difference from predicted BMD and the validity of assuming linear change with time was checked using the mean residuals for each visit number. For spine BMD a chi-squared test showed that the mean residuals were not statistically significantly different from zero. Although statistically significant deviations from linearity were found for the femoral neck and total hip sites the weighted root mean square residuals were small compared with the precision errors. When residuals were binned into histograms a statistical test for skewness was not significant for all three sites. However, a test for kurtosis yielded a statistically significant result for each histogram due to outlying residuals. To determine the standard deviation (SD) of the core gaussian distribution, outliers were trimmed using the method of Melton et al. For lumbar spine BMD outliers with residuals exceeding +/- 3 SD arose mainly from subjects with a body mass index (BMI) >28 kg/m(2) or from subjects who had undergone a large change in BMI during the study. For femoral neck BMD and total hip BMD the outliers were frequently due to inconsistent rotation of the hip. Results for long-term precision calculated from the standard deviation of residuals using the trimmed (untrimmed) data were: lumbar spine BMD, 1.12% (1.65%); femoral neck BMD, 2.21% (2.48%); and total hip BMD, 1.32% (1.57%). These errors were only slightly worse than short-term errors despite changes of DXA scanner during the course of the study. However, obesity may have an adverse effect on precision errors in individual patients and particular care is necessary to ensure reproducible patient positioning for femur scans.