Abstract: To overcome the oscillation and instability in conventional causal inverse-model-based feed forward compensation caused by uncertainties in the manufacturing and assembly of ball-screw-driven systems, a Gaussian-kernel feedforward compensation method for such a system was proposed. The reference trajectories and actual output trajectories of the stage were used as the input and output data for training a Gaussian regression model, and a sliding-window approach was employed to reconstruct the data offline. On this basis, a noncausal finite impulse response function was then used to construct an inverse-model-based feedforward control compensation scheme. Experimental results show that, after application of this feedforward compensation method, the maximum tracking error of the worktable is reduced from 9.4 μm to 5.9 μm. For different trajectories, the maximum tracking error is also reduced by more than 20%. These results indicate that this method significantly improves the motion accuracy of the worktable.