Functional magnetic resonance imaging (fMRI) can be seriously impaired by patient motion. The purpose of this study was to characterize the typical motion in a clinical population of patients in disorders of consciousness and compare the performance of retrospective correction with rigid-body realignment as implemented in widely used software packages. 63 subjects were scanned with an fMRI visual checkerboard paradigm using a 3T scanner. Time series were corrected for motion, and the resulting transformations were used to calculate a motion score. SPM, FSL, AFNI and AIR were evaluated by comparing the motion obtained by re-running the tool on the corrected data. A publicly available sample fMRI dataset was modified with the motion detected in each patient with each tool. The performance of each tool was measured by comparing the number of supra-threshold voxels after standard fMRI analysis, both in the sample dataset and in simulated fMRI data. We assessed the effect of user-changeable parameters on motion correction in SPM. We found the equivalent motion in the patient population to be 1.4mm on average. There was no significant difference in performance between SPM, FSL and AFNI. AIR was considerably worse, and took more time to run. We found that in SPM the quality factor and interpolation method have no effect on the cluster size, while higher separation and smoothing reduce it. We showed that the main packages SPM, FSL and AFNI are equally suitable for retrospective motion correction of fMRI time series. We show that typically only 80% of activated voxels are recovered by retrospective motion correction.