An experimental comparison of model-based joint space position control (JSPC) and Cartesian space position control (CSPC) strategies for the same manipulator, trajectory, sample rate, and dynamic model is presented. It is shown that the achievable tracking performance (in terms of the peak tracking error) and disturbance rejection capability of JSPC was experimentally better than that of CSPC for the same sample rate. High feedback gains for CSPC were found to be unachievable, which we suggest is due to the incompatibility between actuation and control space. The effect of varying the trajectory velocity, and of using a diagonal or full mass matrix in the control torque computation on the performance of the both strategies is also presented. The achievable sample rate on the same control hardware is higher for the JSPC, which leads to yet greater gains in robustness over CSPC. This results from the high computation requirement and the other transformations performed in the CSPC control loop. We conclude that it is possible to choose high feedback gains for JSPC because of the combination of a higher sampling frequency and the compatibility between the actuation space and the control space.