Use of DXA-measured aBMD is the common method to predict osteoporotic hip fractures in clinical settings. However, taking only the changes in aBMD into consideration is not enough to explain the whole variety of low energy fractures. It is deemed essential to develop alternative methods that also reflect the influence of other parameters (e.g. shape of the anatomical structure, load conditions), which are known to be associated with fracture. Development of subject specific FE models is a powerful instrument for investigating bone strength in vivo and, thus, for estimating the risk of fracture. As the mentioned alternative methods need to be adaptable to clinical settings while also being accurate and sufficiently fast for specific tasks, e.g. estimation of proximal femur fracture load, the main aim of this study was to develop a framework that is adaptable to clinical uses and has room for improvement. The presented semi-automatic framework covers development of patient specific FE models based on DXA to predict proximal femur fracture load. Information on the proximal femur shape of individuals were directly derived from DXA by Active Appearance Models (AAM), which detects the object of interest by fitting statistical shape models to the new set of images. To build up AAM, a training data set of DXA scans of 70 proximal femurs was used. Furthermore, 17 DXA scans of the proximal femurs that had been not included in the training set were used as test samples, on which the FE models were developed. To evaluate the effect of segmentation in prediction of proximal femur fracture load, two different cases were considered: proximal femurs that had been segmented using AAM and the same samples with manual segmentation. In order to evaluate the accuracy of AAM, leave-one-out experiments were conducted which provided a point-to-curve error of 1.2470 ±0.6505 (mm) (with 95% confidence). On the other hand, point-to-curve error in segmentation of 17 proximal femurs that were used in the FE analyses was computed as 1.4169 ± 0.7499 (mm) (with 95% confidence). Taking all of the 17 proximal femur samples into account, the fracture loads were estimated to be 3870.9 ± 932.83 (N) for manual segmentation case and 3804.2 ± 850.11 (N) for segmentation case using AAM. A strong correlation was observed between these estimated failure loads (R2 =0.8197). On the other hand, it was noticed that even small errors (e.g. 1.06 mm) in segmentation process might result in larger errors (e.g. 24.1%) in the prediction of fracture load. This work presents the first results obtained with the created framework, which is found to perform sufficiently well compared to its equivalents and is easily adaptable to clinical settings. However, considering the load prediction sensitivity to segmentation, further improvement in the accuracy of the segmentation process is believed to be a vital step for future studies. Such a development might be valuable for the prediction accuracy of proximal femur fracture risk.