Subject-specific finite element (FE) models could improve decision making in canine long bone fracture repair. However, it preliminary requires that FE models predicting the mechanical response of canine long bone are proposed and validated. We present here a combined experimental-numerical approach to test the ability of subjectspecific FE models to predict the bending response of seven pairs of canine humeri directly from medical images. Our results show that bending stiffness and yield load are predicted with a mean absolute error of 10.1% (±5.2%) for the fourteen samples.This study constitutes a basis for the forthcoming optimization of canine long bone fracture repair.