To compare the accuracy of several approaches for defining prevalent vertebral fractures from measurements of vertebral dimensions (morphometry), we measured the lateral dimensions of vertebral bodies of 115 normal premenopausal and 100 postmenopausal women. Of the postmenopausal women two observers agreed that 49 had definite vertebral fractures and 38 were definitely normal. Using these classifications as an independent reference, women were then classified as fractured or normal by several definitions based on vertebral morphometry. No morphometric definition of vertebral fracture agreed perfectly with the consensus classifications. In general, definitions that involved combinations of measurements of anterior (Ha), middle (Hm), and posterior (Hp) vertebral height classified women more accurately than did definitions based on a single measurement or ratio. The Ha/Hp ratio produced many false positives unless it was adjusted for normal variations in the shapes of different vertebral bodies. Definitions of fracture based on a greater than 15% reduction in heights or ratios had higher sensitivity but more false positives than definitions that used a more stringent (greater than 20%) criterion. All morphometric definitions of vertebral fracture separated the post-menopausal women into two groups (fractured and normal) that had significantly (P less than 0.001) different mean spine bone density by quantitative computed tomography. Definitions that had the lowest rates of false positives also produced the largest differences in bone density between those defined as fractured and those defined as normal.