The public Hybrid III family finite element models have been used in simulation of automotive safety research widely. The validity of an ATD finite element model is largely dependent on the accuracy of model structure and accurate material property parameters especially for the soft material. For Hybrid III 50th percentile male dummy model, the femur load is a vital parameter for evaluating the injury risks of lower limbs, so the importance of accuracy of knee subcomponent model is obvious. The objective of this work was to evaluate the accuracy of knee subcomponent model and improve the validity of it.Comparisons between knee physical model and knee finite element model were conducted for both structure and property of material. The inaccuracy of structure and the material model of the published model were observed. Concentrating on the published dummy knee finite element model, mesh refinement was performed, and the foam material in the front of the knee which does not really exist in knee subcomponent product was replaced by dummy skin material. What's more, a series of uniaxial quasi-static compression tests for Hybrid III 50th percentile male dummy knee skin material and rubber were performed. Material parameters of those materials were fitted using uniaxial quasi-static compression test data. By utilizing Hybrid genetic algorithm, two-order Ogden parameters for skin material and Mooney-Rivlin parameters for rubber insert were obtained respectively.After mesh refinement and material model parameters reassignment, the hourglass of the model in the simulation of knee impact calibration test has been well controlled. The modified knee model was validated by knee impact tests under different velocities, and the relative error between the response of simulation and experiment is less than 5%. A more valid knee finite model of Hybrid III 50th percentile male dummy has been established by using a more accurate model structure and more proper material parameters.