Standardized test procedures to evaluate safety performance of cars normally represent real‐world accidents through specific test conditions. It is important, therefore, to understand how performance improvement under such test conditions influences performance in real‐world accidents, which are of course not perfectly represented by such test conditions. The goal of this study is to quantitatively assess how the reduction in pedestrian lower limb injury measures in a standardized subsystem test procedure influences injury measures for pedestrians of different sizes. Three simplified car models, each incorporating two different stiffness characteristics, were used to run impact simulations at 40 km/h against three human finite element (FE) models in different sizes. The direction of change in injury measures was consistent between different pedestrian sizes. The comparison of the change of the peak values of tibia bending moment and knee MCL elongation normalized by the scaled threshold values showed that the change decreases as the size of the pedestrian decreases. It was also found that peak injury measures in different pedestrian sizes can be better estimated when those from a standard anthropometry are scaled under the assumption that contact stiffness between vehicle front‐end structures and a pedestrian does not scale, rather than when elastic modulus was assumed not to scale.