For accurately predicting different fracture patterns of the pelvis frequently observed in pedestrian accidents with SUV/Mini-van, human finite element (FE) models have been developed. Although those models with failure representation can predict occurrence or nonoccurrence of fractures, quantitative estimation of probability of fractures is not possible. For human models without failure representation, typically stress or strain of elements is used for fracture prediction. However, numerous elements must be evaluated when fracture location is not predetermined.This study investigated methodology for accurately predicting probability of pelvic fractures using a minimal number of output parameters. The hood edge and upper and lower parts of the bumper were chosen for representing vehicle fronts. These components were modeled using rigid surfaces with the stiffness of them represented by springs, to constitute 3-component models. FE simulations were run using a human FE model with failure representation and the 3-component models with the location and stiffness of the components varied. Some candidate fracture indices were determined from combinations of displacements, moments and forces. The results with pelvic fractures were used for determining most accurate injury indices based on the normalized variation of the injury index values at fracture. For validating the indices, simplified vehicle models representing the contours of representative SUV and Mini-van were developed using a modeling technique similar to that of the 3-component models. FE simulations were performed using these models and human models with and without failure representation. Pelvic fractures were predicted from eliminated elements for the human model with failure representation, and from the values of the injury index for the human model without failure representation. The results of the fracture prediction were compared for different combinations of simplified vehicle models and stiffness, showing that the injury index determined in this study is capable of accurately predicting various patterns of pelvic fractures.