Areal bone mineral density (aBMD), derived from dual-energy X-ray absorptiometry (DXA) scanners is used routinely to infer bone strength. With DXA hip scans there is growing acceptance of the advantages of also measuring bone structural geometric variables, that complement conventional aBMD to improve understanding of bone modelling, remodelling and processes of metabolic bone disease. However, phantoms for assessing structural geometric variables from DXA scans are not widely available, unlike those for aBMD. This study describes the development of such a phantom, simulating the cortical shell of the human femoral neck, using dental plaster as a material radiologically similar to cortical bone. The mass attenuation coefficient of the dental plaster differed by < 1% from cortical bone, over the relevant energy range. Performance testing was carried out with DXA, to determine accuracy and precision of the phantom structural geometry, using its dimensions and composition as 'gold standards'. Accuracy and precision of cortical structural geometry were poor when measured in a simulated 1 mm-thick osteoporotic cortex (5.5% precision and 50% accuracy errors), but improved with increasing cortical thickness. This study demonstrates the limitations of DXA-based Hip Structure Analysis when applied to femora with thin cortices, and indicates improvements in the design of a phantom to better simulate such cortical structures.