Or realignment in SPM terms.
Choose the "Realign" button, and select the "Realign" option from the small drop-down menu. Set the number of subjects (maybe 1 here), and when asked for number of sessions, give the number of functional runs for this subject. If you have done slice timing already (see SliceTiming), select all the slice timing corrected images, for each functional run (session) in turn, as prompted. If you are doing motion correction before slice timing, select the so-far unprocessed images. The first image in the first session should be an image you are happy with, i.e. not one of the images you are going to discard. Choose "Coregister and reslice", "Create mean image only" when prompted to do so. Wait and wait. You now have a new mean*.img; the *.mat files for all the images you selected will have been adjusted to reorient the images to the first image that you selected. For each run (session), in the directory containing the files for that run, there will also be a text file called "rp_ [first_image_name].txt. This text file has six columns of numbers, and the same number of rows as there were scans in the run; the columns contain respectively the x y z translations, and x y z rotations to reorient the matching image to the first image in that run (remember the .mat files now contain the transformations necessary to reorient the images to the first image in the first run). You may need the rp_ text file in the statistics later.
If you want to do an analysis on the data without doing spatial normalisation or you want to apply EPI undistortion, you will also need to reslice the *img series that you are calculating the realigment parameters for. In that case, instead of choosing "Create mean image only", you would choose "All images + Mean image". You will have to wait even longer of course, and you will get a new set of files with an "r" appended to them, which have been resliced to match the first image in the first run.
You have now done spatial motion correction. Look at the motion parameter display in the graphics window to check for large movements during the functional runs (movement between the runs is not usually a problem). Typical translations within a run are less than a couple of mm. Note that small movements do not at all mean that you will not have movement artefacts. Anecdotal evidence suggests that small fast movements can have as much effect on the time course as large slow ones.
Realign and unwarp or just realign?
Jesper Anderson wrote a tool that models how the brain changes in shape with small motions due to distortion in the image. If there are substantial distortions in your images this may be useful. However, unless you are testing patients with clips, or have some other reason to expect distortions, this tool is probably not vital.