Permanent-magnet step motors offer several advantages such as high efficiency, high power density, high torque-to-inertia ratio, and excellent durability and serviceability, as well as the absence of external rotor excitation and windings. The nonuniformity in the developed torque due to the nonsinusoidal flux distribution in the airgap is, however, the major obstacle in achieving global high-precision position tracking. When the position reference profile is a periodic signal of known period, such an obstacle may be however overcome by using recent learning control techniques, which require neither high gains in the inner speed/position control loops nor resetting procedures. An experimental comparison of two different recently designed learning position controls (“adaptive” and “iterative”) is, for the first time, carried out with reference to the same low-speed robotic application. Benefits and drawbacks of the two learning approaches are analyzed in detail.