Objective <p>T2*-weighted acquisitions used for functional neuroimaging suffer from signal losses in the presence of field inhomogeneities due to through-slice dephasing that can be ameliorated with so-called z-shim gradient pulses in the slice direction. Determining the z-shim settings from the negative linear field component in the slice direction seems to be reasonable but the performance of this approach in the presence of non-linear field variations is unclear.</p> Materials and methods <p>Here, numerical simulations are performed for different through-slice field variations and echo times&#xa0;(TE) in order to determine the optimum z-shim setting that maximises the signal amplitude.</p> Results and discussion <p>They demonstrate that the optimum z-shim compensation gradient field, i.e. the gradient pulse time integral divided by TE, differs from the negative linear field component and depends on TE. The results also indicate that non-linear field inhomogeneities may cause a spatial weighting of the signal within the slice that is modulated with the z-shim. Furthermore, it is shown that these findings also hold for a region-of-interest composed of voxels with purely linear field variations that differ between voxels. Thus, deriving z-shim values from the linear field component(s) of a voxel or region-of-interest will, in general, not provide the maximum signal amplitude.</p>