Clinical questionnaires and dental radiographic findings have both been suggested as methods of identifying women at risk of having osteoporosis and who might benefit from bone densitometry. The aim of this study was to measure the diagnostic accuracy of a combination of mandibular cortical width (MCW) measured from dental panoramic radiographs (DPRs) and the osteoporosis index of risk (OSIRIS) in the diagnosis of osteoporosis. 653 women (age range 45-70 years, mean age 54.95 years) in four European centres underwent standardised dual X-ray energy absorptiometry (DXA) to provide reference data on osteoporosis status. Each subject was interviewed to derive OSIRIS scores and underwent DPR examination. MCW was measured directly by five observers. Receiver Operating Characteristic (ROC) curve analysis was used to calculate sensitivities and specificities of the clinical and radiographic tests for the diagnosis of osteoporosis. 512 (78.4%) of the study population were classified as having normal BMD and 141 (21.6%) as having osteoporosis. Using ROC analysis, OSIRIS gave a ROC curve area (A(z)) of 0.838, with a sensitivity of 70.9% and a specificity of 79.5% at a diagnostic threshold of <or=+1. MCW on DPRs gave Az values for the five observers ranging from 0.71 to 0.78, providing sensitivities between 41.0% and 59.6% and specificities of between 81.8% and 90.3% at a diagnostic threshold of a 3 mm MCW and sensitivities between 94.2% and 99.3% and specificities of between 9.8% and 23.7% at a diagnostic threshold of a 4.5 mm MCW. Inter-observer repeatability was less than 2.15 mm for 95% of subjects. Combining clinical and radiographic tests had the effect of improving specificity at the expense of a fall in sensitivity. Diagnostic thresholds for MCW and OSIRIS can be chosen to provide the sensitivity and specificity combination that best suits locally determined needs. However, the addition of OSIRIS as a stepwise 'follow-up' test to radiographic assessment of MCW should only be performed if the aim is to have a test for which the highest achievable specificity is desired.