Morphometric detection of vertebral fractures requires the use of reference normal values. In previous studies, normal values were obtained in young healthy adults or in adults of various ages in whom abnormal radiographs were excluded using trimming algorithms. We established four sets of normal values defined according to different approaches. We used a group of 102 young healthy women with normal spine radiographs and a population-based cohort of 260 women 51 to 85 years old (EVOS study) to select those with normal spine radiographs based on the semiquantitative method of Genant, the trimming algorithm defined by Melton, and the trimming algorithm defined by Black. Using the four sets of reference data, we applied two diagnostic criteria, the mean - 3SD and the 0.85 x mean to detect prevalent vertebral fractures in two groups of women: the population-based cohort of 260 women recruited for the EVOS study and 176 osteoporotic women 54 to 88 years old with at least one vertebral fracture. In the population-based cohort, the number of vertebral fractures varied from 53 to 57 for the 0.85 x mean cutoff and from 93 to 115 for the mean - 3SD cutoff. In the osteoporotic women, the number of vertebral fractures varied from 255 to 273 for the 0.85 x mean cutoff and from 372 to 404 for the mean - 3SD cutoff. In both groups, the number of vertebral fractures detected with the 0.85 x mean cutoff was close to the number detected by the semiquantitative method of Genant, which does not require a reference population. For the 0.85 x mean cutoff, agreement of diagnosis using different reference values was excellent (kappa = 0.96-0.99). For the mean - 3SD cutoff, agreement of vertebral fracture detection based on the four reference sets was good (kappa = 0.79-0.93). Agreement of diagnosis was lower when two cutoffs were compared using the same set of reference values (kappa = 0.60-0.81). In conclusion, reference data derived using different approaches give similar rates of prevalence of vertebral fractures. Compared to the semiquantitative assessment, the prevalence of vertebral fractures is similar for the 0.85 x mean cutoff, whereas it is considerably higher for the mean - 3SD cutoff.