A recent analysis has found that during treatment with denosumab, women attaining higher bone densities (BMD) are less likely to have incident fractures. We have reexamined this important question using data from our recent trial of zoledronate in osteopenic women. One thousand women randomized to treatment with zoledronate were followed for 6 years. Of those, 122 sustained fragility fractures during follow‐up. Baseline age, nonvertebral fracture history, total hip BMD, and calculated fracture risk were all significantly different between those who had fractures during the study and those who did not. BMDs achieved during the study were higher in those without incident fractures. However, achieved BMDs were very closely related to baseline values (r = 0.93, p < 0.0001). The increase in BMD during zoledronate treatment was not different between those who had incident fractures and those who did not (0.15 < p < 0.78), and change in BMD was not predictive of fracture (univariate logistic regression analysis). Stepwise regression analysis of all baseline variables showed the best independent predictors of fracture to be age (odds ratio [OR] = 1.08, 95% confidence interval [CI] 1.04–1.13, p = 0.0003), baseline spine BMD (OR = 0.81, 95% CI 0.67–0.96, p = 0.016), and history of nonvertebral fracture (OR = 1.69, 95% CI 1.06–2.69, p = 0.028). Addition of change in BMD to this model did not improve its predictive power. If changes in BMD were included in the stepwise regression analysis of baseline variables, they did not emerge as significant predictors of fracture. It is concluded that age, fracture history, and baseline BMD determine the risk of new fractures. Differences in achieved BMD between those who do or do not fracture arise from the close relationship between baseline and achieved BMDs. These findings suggest that targeting any particular BMD during treatment is unlikely to be a useful or valid strategy.