This paper discusses the use of acceleration measurements to improve the performance and robustness of controllers for mechanical systems. To use acceleration signals there are at least two approaches: direct use in a feedback loop to improve the tracking error, and indirect use by an observer to improve the estimates of position and speed. Several proposals for the use of the acceleration in a feedback loop, resulting in slightly different controllers, are discussed. The design of the controllers for the experimental system, a flexible multiple degrees-of-freedom XY-table, uses a simplified two degrees-of-freedom model. The observer is of the predictive type to compensate for the time delay in the implementation, and its design is based in part on Kalman filter theory. Simulations and experiments show that both acceleration feedback and an acceleration assisted observer can improve the performance of the control system, but the robustness does not change significantly. A combination of both approaches did not give any improvement in the experiments, but some in the simulations. Disturbances in the acceleration signal (especially motor torque ripple), phase lag introduced by the signal processing equipment, time delay caused by the sampled data implementation of the controller and observer, and the non-colocated position and acceleration sensors are believed to be limitations for the usefulness of the acceleration signal. A performance improvement up to a factor of 1.5 was possible in the author's application, so to consider the use of acceleration measurements in tracking control is recommended.