We investigate the effect of specimen geometry on the ultimate tensile strength of cortical bone measured by a tensile test. This article is motivated by the fact that there is no clear consensus in the literature on a suitable specimen shape for cortical bone testing. We consider three commonly used tensile test specimen shapes: strip, dumbbell with sharp junctions, and dumbbell with rounded junctions. We conduct this study computationally, using a finite element method, and experimentally by testing porcine femurs. Our results show that local stress concentration factors in the specimen lead to reduced values in the measured tensile strength. The higher the stress concentrations are, the lower is the measured strength. We find that the strip specimens are not a good choice due to high stress concentrations. For the same reason, dumbbell specimens with sharp junctions between the grip and gage sections should also be avoided. The dumbbell shaped tensile test specimens with an arc transition and a maximized radius of fillet are a better choice because such geometry lowers stress concentrations.