Autocorrection is an adaptive motion correction algorithm that does not require an in vivo measurement of the motion record. A novel method for ensuring convergence of this algorithm when motion is severe is presented. A limited number of navigator echoes are acquired before the imaging sequence to obtain a "snapshot" of the object. Phase differences between the navigator and image k-space data are used as an estimate of motion-induced phase shifts in the image, followed by autocorrection. In phantom data a six-fold reduction in computation time compared to autocorrection alone was realized. These results indicate that this navigator/autocorrection combination may be useful for reducing motion artifacts and computation time for MR exams when motion along the image phase encoding axis is severe.