Noticeable differences exist between treadmill and overground walking; kinematics, kinetics, and muscle activation patterns differ between the two. Many previous studies have attributed the differences to changes in visual information, air resistance, and psychological effects such as fear. In this study, we demonstrate that no treadmill serves as an inertial frame of reference. Considering the linear momentum principle, the finite sampling rate of the controller, and the limited power of the treadmill motor, we predict that 1) the error of the treadmill speed periodically varies depending on the locomotion phase and 2) this non-ideal behavior becomes more evident as the locomotion speed or the weight of the walker increases. Experimental observation confirmed our predictions by quantifying the variation of the actual treadmill belt speed and the ground reaction force in the anterior–posterior direction for different locomotion speeds and subject weights. These results emphasize a need for design criteria like the minimum sampling rate and the minimum motor power that treadmill locomotion studies should consider.