Movement-related effects in realigned fMRI timeseries can be corrected by regression on linear functions of estimated positional displacements of an individual subject's head during image acquisition. However, this entails biased (under)estimation of the experimental effect whenever subject motion is not independent of the experimental input function. Methods for diagnosing such stimulus-correlated motion (SCM) are illustrated by application to fMRI data acquired from 5 schizophrenics and 5 normal controls during periodic performance of a verbal fluency task. The schizophrenic group data were more severely affected by SCM than the control group data. Analysis of covariance (ANCOVA) was used, with a voxelwise measure of SCM as a covariate, to estimate between-group differences in power of periodic signal change while controlling for variability in SCM across groups. Failure to control for SCM in this way substantially exaggerated the number of voxels, apparently demonstrating a between-group difference in task response.