It has been proposed that vertebral dimensions be used to objectively identify vertebral fractures, permitting standardization of methodology for comparisons between studies. In this report, we evaluate the ability of various vertebral dimensions and ratios to identify "abnormal" vertebrae. As no "gold standard" exists for prevalent vertebral fractures, we examined the ability of cross-sectional dimensions (at a single point in time) to detect fractured vertebrae that had been identified from changes in dimensions compared with previous radiographs. Theoretically, a cutoff of 3 SD below the mean will rarely misclassify normal vertebrae as fractured (specificity = 99.9%). However, we found that this cutoff correctly identified only about 70% of the incident fractures. A less stringent criterion (2 SD below the mean; theoretical specificity = 97.7%) identified about 85-90% of true fractures. Dividing by stature or other vertebral heights sometimes yielded marginal improvements in the ability of the anterior or posterior height dimensions to diagnose fractures. The results suggest that the true fracture prevalence may sometimes be substantially higher than suggested by cross-sectional vertebral measurements.