Patient motion during the acquisition of a magnetic resonance image can cause blurring and ghosting artifacts in the image. This paper presents a new post-processing strategy that can reduce artifacts due to in-plane, rigid-body motion in times comparable to that required to re-scan a patient. The algorithm iteratively determines unknown patient motion such that corrections for this motion provide the best image quality, as measured by an entropy-related focus criterion. The new optimization strategy features a multi-resolution approach in the phase-encode direction, separate successive one-dimensional searches for rotations and translations, and a novel method requiring only one re-gridding calculation for each rotation angle considered. Applicability to general rigid-body in-plane rotational and translational motion and to a range of differently weighted images and k-space trajectories is demonstrated. Motion artifact reduction is observed for data from a phantom, volunteers, and patients.