The aim of the study was to validate a non-invasive method allowing a semi-automatic cortico-medullar index (CMI) computation from standard radiographs. Fifty-five radiographs of the non-dominant radius of prepubertal girls (aged 10.12+/-1.47 years) were scanned (100 microm scanning resolution) by the same observer. To detect the grey level peaks, we used the Deriche's recursive filter applied to the grey level gradients and we defined the cortical areas by thresholding (iterative contour detection). The CMI was calculated as cortical areas divided by the total bone area. The reproducibility was tested through the analysis of ten radiographs scanned twice, on 5, 10, 15, 20, 30 and 40 lines at each side of a reference line located at the distal third part of the radius. We used the root mean square coefficient of variation (RMS-CV) as verification. Dual-Energy X-ray (DXA) was used for densitometry measurements of the whole body, and non dominant radius (mid-, distal third and ultra distal). The result attaining the best reproducibility concerned the 15-line position (Mean CMI=0.58+/-0.005 with a 1.36% RMS-CV). CMI correlated significantly with Bone Mineral Density (BMD) (r=0.40, 0.33 and 0.50, P<0.01, respectively, at the whole body, mid- and third distal of the radius), Bone Mineral Content (BMC) (r=0.32, 0.36, 0.34 and 0.35, respectively, at the whole body (P=0.01), mid- (P=0.007), third distal (P=0.01) and ultra distal of the radius (P=0.01)) but not with BMD at the ultra distal of the radius (r=-0.01, P=0.9). This semi-automated new method has been easily implemented, reproducible, and therefore, could be useful tool for the bone CMI computation.