Hi Roberto, Is this based on a higher-level analysis? The first thing is to make absolutely sure that the statistical model (i.e., GLM) in whatever software package that you're using to conduct the analysis of the ROI values is identical to the GLM specification that you used to conduct the whole brain analysis in FSL. That includes having the identical effects in the design matrix (all the same main effects, interactions, and any other covariates), and making sure that any demeaning of variables is the same as well. Claims of putative inconsistencies like this are a common post to various lists (FSL, SPM, FreeSurfer), but I suspect that the issue is usually that the model used for the ROI analysis doesn't match what was used for the whole brain analysis. cheers, -MH On Fri, 2011-05-06 at 11:17 +0100, Roberto Ferreira wrote: > Hi list, > > I've conducted a whole brain analysis and I've got a reliable difference between condition A and condition B (cluster corrected) with z=3.98 for highest peak within the cluster. Then I decided to run a ROI analysis, so I drew a mask of 7x7x7 mm to look at the percent signal change within the masked data. However, I didn't get any significant difference between the conditions. Is it common to find inconsistency between whole-brain and ROI analyses? I'm a beginner in fMRI, so I'm a bit lost here. > > I would really appreciate some ideas on this matter. > > Many thanks, > > Roberto