Hi Anderson, Thank you for this explanation. I think I entered everything as you described. I have 10 functional networks as input, 10 design matrices (1 per input) and 10 voxelwise EVs (1 per input/matrix), and column no. 18 in the design matrix is the voxelwise EV position. My PALM command looks like this: palm -i IC00.nii -i IC01.nii -i IC02.nii -i IC03.nii -i IC04.nii -i IC05.nii -i IC06.nii -i IC07.nii -i IC08.nii -i IC09.nii -o Palm -m mask.nii -designperinput -d IC00.mat -d IC01.mat -d IC02.mat -d IC03.mat -d IC04.mat -d IC05.mat -d IC06.mat -d IC07.mat -d IC08.mat -d IC09.mat -t contrasts.con -evperdat EV_IC00.nii 18 1 -evperdat EV_IC01.nii 18 2 -evperdat EV_IC02.nii 18 3 -evperdat EV_IC03.nii 18 4 -evperdat EV_IC04.nii 18 5 -evperdat EV_IC05.nii 18 6 -evperdat EV_IC06.nii 18 7 -evperdat EV_IC07.nii 18 8 -evperdat EV_IC08.nii 18 9 -evperdat EV_IC09.nii 18 10 -eb EB.grp -n 2000 -corrmod -fdr -T -save1-p Without the -evperdat arguments, there are no problems. But when I include the voxelwise EVs in the command line, I receive the following error: error: palm_takeargs: A(I): index out of bounds; value 11 out of bound 10 error: called from: error: /usr/local/LKeb/FSL/PALM/palm_takeargs.m at line 1796, column 56 error: /usr/local/LKeb/FSL/PALM/palm_core.m at line 32, column 11 error: /usr/local/LKeb/FSL/PALM/palm.m at line 80, column 1 Any idea on how to handle this? Furthermore, since you explained that the voxelwise EV is inserted in the matrix, I wonder whether it is really necessary to select the specific 4D file manually for each design. Since my designs only differ with regard to this voxelwise EV, could I just use one design (with one random voxelwise EV) for all 10 inputs and refer to the correct EV with -evperdat? I'm also concerned about the EBs that I use. We have 1 group of subjects. Each subject is present at 2 different study days (drug vs. placebo). On both days, 5 fMRI scans are repeatedly acquired (1 pre, 4 post treatment). We are interested in the treatment effect (paired t-test with contrasts 1 and -1 for one EV). Time variance (#EVs = n-1 time points) and random intercepts (1 EV per subject) are modelled as regressors of no interest. The design is balanced (there are no missings). I believe that multi-level block permutation might be appropriate, or is it not allowed to permute at all, and should I use -ise? Or is it sufficient for this (within) contrast to just permute within blocks (1 block per subject)? For instance, for 2 subjects, the EB that I currently use looks like: -1 1 -1 1 -1 1 -1 1 -1 1 -1 2 -1 2 -1 2 -1 2 -1 2 -1 1 -1 1 -1 1 -1 1 -1 1 -1 2 -1 2 -1 2 -1 2 -1 2 I'm not sure about the right approach. Looking forward to your advice, Best, Bernadet