Hi Marco, Could you post here your design.mat and design.con? It's just to see if there isn't anything strange or unexpected going on that could otherwise explain these high values. About the relationship between the statistic and the p-values: using a parametric method, yes, a t-stat of about 2 would have a p about 0.05 (in fact, 0.03, but close). This only happens, however, if various assumptions about the data are valid. The parametric methods were developed at a time in which computers didn't exist, so theoretical approximations and asymptotic behaviours of random variables had to be investigated, and for that, assumptions were introduced. We don't need any of that anymore, and randomise uses methods that have fewer assumptions and are more generalisable. It computes exact p-values, even for distributions that can have weird shapes for which no theoretical approximation exist. A t-statistic that may seem "high" from a parametric perspective (so with low parametric p-value) may in fact be a common value in your experiment using your data, hence a not so low p-value, which indeed you are observing. This and other issues are discussed in the randomise papers: this<http://dx.doi.org/10.1002/hbm.1058>and this <http://dx.doi.org/10.1016/j.neuroimage.2014.01.060>. All the best, Anderson 2014-04-29 10:59 GMT+01:00 SUBSCRIBE FSL Anonymous <[log in to unmask]> : > Thanks for the quick answer Anderson. > > If I understand correctly, the high values of t-stat I obtain in my > analysis might be genuine. however I still do not understand the link > between these t-values and the p-values obtained in the uncorrected p stat > images. > > In fact, if my sample is n=20, I would expect an uncorrected t-stat just > higher than 2 to be significant at a 0.95 level, right? So can you help me > to understand why if i threshold the t-stat image at 2 almost all voxels > are significant whereas when I threshold the p stat image at 0.95 almost > none of them is significant? how do i choose the threshold in the t-stat > image that corrsponds to a 0.95 level of significance? > > thanks for your help. >