A technique for producing geometrically accurate magnetic resonance images (MRIs) with undistorted intensity in the face of high levels of static field inhomogeneity arising from either source is presented. The technique requires the acquisition of two images of the same object with altered gradients. On the basis of a knowledge of these gradients it employs an automatic postprocessing step that exploits some invariant characteristics of the distortions to produce a rectified image from the two acquired images. No phantom imaging is involved and no operator interaction is required. The technique is theoretically justified and compared to other techniques, and experimental results that show that the technique works are presented. The improved accuracy in geometry and intensity may improve reliability of stereotactic surgery, may enhance the feasibility of both clinical and industrial imaging via external fields, and may increase the resolution of microscopic imaging.