Subject motion present during the time course of functional activation studies is a pervasive problem in mapping the spatial and temporal characteristics of brain activity. In functional MRI (fMRI) studies, the observed signal changes are small. Therefore, it is crucial to reduce the effect of subject motion during the acquisition of image data in order to differentiate true brain activation from artifactual signal changes due to subject motion. We have adapted a technique for automatic motion detection and correction which is based on the ratio-variance minimization algorithm to the fMRI subject motion problem. This method was used for retrospective correction of subject motion in the acquired data and resulted in improved functional maps. In this paper we have designed and applied a series of tests to evaluate the performance of this technique which span the classes of image characteristics common to fMRI. These areas include tests of the accuracy and range of motion as well as measurement of the effect of image signal to noise ratio, focal activation, image resolution, and image coverage on the motion detection system. Also, we have evaluated the amount of residual motion remaining after motion correction, and the ability of this technique to reduce the motion-induced artifacts and restore regions of activation lost due to subject motion. In summary, this method performed well in the range of image characteristics common for fMRI experiments, reducing motion to under 0.5 mm, and removed significant motion-induced artifacts while restoring true regions of activation. Motion correction is expected to become a routine requirement in the analysis of fMRI experiments. © 1995 Wiley-Liss, Inc.