Thanks for the answer Steve, On Wednesday 24 January 2007 18:13, Stephen Smith wrote: > Hi - I think your preprocessing steps are ok. Which output from dtifit are > you feeding into randomise, the FA data? I am looking at FA, MD, L1 (longitudinal diffusion) and avg(L2, L3) (transverse diffusion). > > A t-test model in randomise would be sensible if you were comparing a set > of zero-error FA images with a set of non-zero (but equivalent) error FA > images (eg generated from a set of different but similar "badness" original > added rotations. Is this what you are doing? Yes, that's what I do. E.g., I generate 10 series with RMS of x mm (this should be non-zero error) and compare it with another 10 series with zero error. > > The simplest output from randomise to look at is the _vox image, ie > uncorrected voxelwise p- values for the t-test. I will re-check intermediate results of my preprocessing to make sure everything is as it should be. Have a nice day. Martin