Prospective and cross-sectional studies have demonstrated that bone mass predicts fracture risk. However, most prospective studies have been limited to a few years of follow-up. We investigated the long-term associations of bone mass with vertebral fractures using longitudinal data collected from more than 500 postmenopausal Japanese-American women in the Hawaii Osteoporosis Study. New vertebral fractures were identified during an average of 2.7 years between 1992 and 1995. Short-term fracture prediction was evaluated using bone mass (spine, calcaneus, distal radius, and proximal radius) measured at the beginning of follow-up. Long-term prediction was evaluated using bone mass measured before the follow-up period (11 years earlier for nonspine bone mass and 8 years earlier for spine). All four bone mass measurements were significant predictors of vertebral fractures identified during the subsequent 2.7 years (short-term prediction), with odds ratios (ORs) ranging from 1.5 to 1.9. The ORs for long-term prediction were slightly lower in magnitude, but the confidence intervals overlapped the short-term ORs considerably, suggesting that both long-term and short-term associations are similar in magnitude. Furthermore, cross-sectional analyses based on bone mass measurements performed at the end of follow-up (after fractures had occurred) yielded results similar to those based on prospective data (bone mass measured prior to fractures), suggesting that the relatively quick and inexpensive cross-sectional studies are useful for preliminary evaluations of new bone mass measurement techniques. The results suggest that bone mass measurements made up to 11 years earlier can predict vertebral fractures almost as well as measurements made more recently.