Recent studies have shown that a low bone mineral density (BMD) is associated with a higher risk of mortality. Most studies have investigated this relationship in women only and presented their risk estimates per standard deviation change in BMD. However, when using this approach, a BMD threshold might be missed when relative risks are presented in the traditional manner. Therefore, in this study our aim was to model the relation between BMD and all-cause mortality. In the Rotterdam Study, follow-up was complete for 5819 men and women aged > or =55 years for whom BMD data were available. During an average follow-up of 5.4 years, 399 men and 317 women died. We calculated BMD Z scores using measurements performed at the femoral neck. Cox proportional hazards regression was used to fit the model. An average BMD, reflected by a Z score = 0, was used as the reference. For women, no significant relationship between BMD and overall mortality was observed. For men, however, a cubic model best fitted the relationship under study, also after adjusting for age and body mass index (BMI). The risk of mortality increased when BMD was below average. Similar results were found when separate curves were made for diabetics and nondiabetics, smokers (ever or never), and tertiles of BMI. Excluding subjects who had suffered hip fractures, or adjusting for the number of drugs used and for lower limb disability, essentially did not change results. This suggests that low BMD is not mainly due to morbidity and impaired mobility in our cohort, which makes this a less likely explanation for the observed relation with mortality. The results of our study suggest that, in men, a nonlinear relationship between BMD and mortality exists, which is independent of comorbidity, whereas, in women, no significant relationship was observed.