Geometric distortions caused by magnetic susceptibility variations in the underlying medium can severely corrupt the image. A novel correction method is proposed, which uses the prior knowledge of the analytic point spread function (PSF) of the used imaging sequence and a map of the underlying field inhomogeneities. From this input, a PSF operator can be devised and applied to correct the image by performing a deconvolution. Regularization techniques are used to improve and stabilize the outcome. A significant reduction in geometric distortions is demonstrated for human brain images as well as some advantages over existing correction methods.