A new method is proposed for reducing the artifacts produced in diffusion-weighted imaging. When data are acquired using multiple shot echo planar acquisitions, conventional reconstruction methods produce artifactual images as a consequence of diffusion weighting and small amounts of bulk motion of the subject. If the amount of motion can be determined, it is possible to correct the data before reconstruction, which removes the artifact. A method for estimating the motion from the acquired data has been developed and evaluated. This method assumes that ghost image effects will be minimized when motion has been correctly compensated. By considering the amount of signal in the background of the image, appropriate corrections to the data can be made, and the accuracy of the motion compensation may be estimated. This technique has been evaluated by computer simulation, and its performance has been demonstrated in a phantom and humans with both two- and four-shot echo planar acquisitions and using both "mosaic" and "interleaved" sampling schemes.