The purpose of this study was to develop injury risk functions for dynamic bending of the human femur in the lateral-to-medial and posterior-to-anterior loading directions. A total of 45 experiments were performed on human cadaver femurs using a dynamic three-point drop test setup. An impactor of 9.8 kg was dropped from 2.2 m for an impact velocity of 5 m/s. Five-axis load cells measured the impactor and support loads, while an in situ strain gage measured the failure strain and subsequent strain rate. All 45 tests resulted in mid-shaft femur fractures with comminuted wedge and oblique fractures as the most common fracture patterns. In the lateral-to-medial bending tests the reaction loads were 4180 ± 764 N, and the impactor loads were 4780 ± 792 N. In the posterior-to-anterior bending tests the reaction loads were 3780 ± 930 N, and the impactor loads were 4310 ± 1040 N. The difference between the sum of the reaction forces and the applied load is due to inertial effects. The reaction loads were used to estimate the mid-shaft bending moments at failure since there was insufficient data to include the inertial effects in the calculations. The resulting moments are conservative estimates (lower bounds) of the mid-shaft bending moments at failure and are appropriate for use in the assessment of knee restraints and pedestrian impacts with ATD measurements. Regression analysis was used to identify significant parameters, and parametric survival analysis was used to estimate risk functions. Femur cross-sectional area, area moment of inertia ( I ), maximum distance to the neutral axis ( c ), I/c , occupant gender, and occupant mass are shown to be significant predictors of fracture tolerance, while no significant difference is shown for loading direction, bone mineral density, leg aspect and age. Risk functions are presented for femur cross-sectional area and I/c as they offer the highest correlation to peak bending moment. The risk function that utilizes the most highly correlated (R 2 = 0.82) and significant (p = 0.0001) variable, cross-sectional area, predicts a 50 percent risk of femur fracture of 240 Nm, 395 Nm, and 562 Nm for equivalent cross-sectional area of the 5 th percentile female, 50 th percentile male, and 95 th percentile male respectively.