BMD is commonly obtained with DXA, but this is confounded by the length and composition of tissues that the X-ray must traverse. Subpixel enhancement of nonuniform tissue (SPENT) is a novel MRI technique that can provide (direction specific) information based on the subvoxel structural uniformity of a sample. We hypothesized that the SPENT signal would be related to BMD. This hypothesis was tested using (1) 2D computer simulation of a simplified bone structure and (2) in vitro experiments. Simulation results suggested that a resolution of 610-800 microm was required for SPENT to be correlated well with the simulated bone volume fraction (BVF) and, at this resolution, a modest signal-to-noise ratio (SNR > 5) was required for reasonable data quality. For the experiments, 15-mm(3) human trabecular bone samples were used (1) to quantify BMD (through both physical measurement and DXA) and (2) to perform MRI on a 7T system. Standard and SPENT images were obtained. Normalized SPENT (NSPENT) images were calculated by pixel-by-pixel division of the SPENT images by the standard proton density images to remove any dependence on proton density and coil uniformity from the SPENT images. The average NSPENT values were determined over the sample volume and compared with the reference BMD measurements. Each of the individual NSPENT directions was highly correlated with BMD (x-NSPENT, R (2) = 0.73, p < 0.001; y-NSPENT, R (2) = 0.76, p < 0.001; z-NSPENT, R (2) = 0.89, p < 0.001). With all three individual NSPENT directions combined, the correlation with BMD was found to be the highest (xyz-NSPENT, R (2) = 0.95, p < 0.001). The results suggest that the SPENT technique can provide a noninvasive measure of BMD at resolution and SNR levels achievable in vivo.