The fragility of trabecular bone depends not only on the amount of bone but also on its architecture. In order to assess fragility of bone, describe changes due to age, and monitor effect of disease or treatment, it is necessary to model the physical properties of trabecular bone architecture. An important feature of bone architecture is the degree of anisotropy (DA). Estimates of DA may be obtained from computed tomography data by characterizing orientation in images. Widely used image descriptors for estimating orientation in this setting include mean intercept length (MIL), star volume distribution (SVD), star length distribution (SLD), and line fraction deviation (LFD). In this study, estimates of DA computed via each of these image descriptors are compared on synthetic images for various combinations of trabecular thickness, separation and number. Estimates of DA are also computed for real images representing different stages of aging. It is found that estimates of DA vary substantially depending on the choice of image descriptor. In particular, the MIL tends to underestimate DA.