Metabolic bone diseases such as osteoporosis usually cause a decrease in bone mass and a deterioration of bone microarchitecture leading to a decline in bone strength. Methods to predict bone strength in patients are currently based on bone mass only. It has been suggested that an improved prediction of bone strength might be possible if structural changes are taken into account as well. In this study we evaluated which structural parameters (other than bone mass) are the best predictors for changes in bone mechanical properties of the human radius after different bone atrophy scenarios and whether the original strength of the affected bone can be recovered if bone loss is restored by thickening of the remaining structures. To answer these questions, a human radius was measured with a microcomputer tomography scanner to extract the full three-dimensional architecture of the distal radius at an isotropic resolution of 80 microm. Eight models with modified bone architecture were created and the mechanical variations due to these modifications were studied using microfinite element (micro-FE) simulations. In four models mass was lowered by 20%, either by reducing cortical thickness, trabecular thickness, or number of trabeculae or by overall thinning of structures. In the other four models bone mass was restored to the original value using a trabecular bone thickening procedure. The micro-FE analyses revealed that most load was carried by the cortical bone. For this reason, bone strength was affected most in the reduced cortical thickness model. For the same reason, the trabecular bone atrophy scenarios, all of which affected bone strength in a very similar way, resulted in less dramatic bone strength reduction. The restoration of bone mass did not recover the original bone strength. These findings demonstrate that the importance of different parameters for the prediction of bone strength also depends on the mechanical loading. This could explain why results of earlier studies on the importance of structural parameters can be inconsistent and site-dependent.