Here's the pathway I took. I'm unsure if this is ok, considering the way FSL thresholds results. 1. 3 Groups: Schizophrenia, Schizophrenia Relatives, Controls 2. Contrast is Red Cue vs. Green Cue (RCvsGC) 3. Group level analysis: Controls (1) > Relatives (-.5) > Schizophrenia (-.5) 4. Results in RCvsGC in Controls > Relatives > Schizophrenia with cluster threshold of 2.3. 5. Use fslmaths to intersect the cluster_threshold_mask_stat with inferior frontal gyrus (Harvard Oxford) and create mask. 6. Run uncorrected main effect of RCvsGC in each of 3 groups. 7. Use featquery to mask the uncorrected main effect of RCvsGC in each group with the mask created in step 5. 8. Extract mean time series (which I understand is each subject mean %change in activation when you click on mean time series in featquery results) 9. Compare the 3 groups using t.tests, anova, etc.... to establish non-overlapping error bars (or lack there of). My main question would be in relation to using that specific interaction(RCvsGC) and contrast(1, -.5, -.5) to create a mask. Is this principled. Secondly can I use that mask to pull beta weights associated with uncorrected main effects, because the group level was already cluster thresholded? Edward H. Patzelt University of Minnesota