Computational fracture analysis has become a growing branch of orthopedic research. Particularly, the associated methods provide reliable tools for the analysis of 3D CT-based models of bone. This paper reports the results of such analyses for 15 human femora (healthy and osteoporotic) under different loading orientations (85 different analysis cases). A new method was developed for the calculation of the density distribution in the models from ordinary clinical CT images without calibration phantom. This method, along with a strain-energy-based linear finite element (FE) analysis scheme, was used to predict the fracture strength and pattern of 10 cadaveric femora, for which the mechanical testing results and calibrated FE models were already available. The very good agreement and consistency between different sets of results showed the reliability and accuracy of the new density calibration method, as well as the linear analysis scheme. Accordingly, the method was applied to five new clinical images, gathered from two clinics that used different scanners with different protocols. The strength and fracture pattern of each one of these specimens were analyzed under 15 different loading conditions. A consistent behavior was found for variation of the fracture load and pattern of all specimens with the loading orientations, while very clear contrasts were observed between the strength amplitudes of healthy and osteoporotic specimens. The proposed methods can be easily applied to ordinary daily (even archived) clinical CT scans to conduct fast and reliable fracture analysis of human femora for general bone research and opportunistic studies of osteoporosis and trauma.