Dear Mark,

> I analysed a contrast using randomise and confined the analysis within a mask. In another strategy, this mask was entered as pre-threshold masking in post-stats of feat. It turned out that there was significant activation in randomise method but no activation in pre-threshold masking method.
>
> I know randomise used permutation and feat used GRF, but is this discrepancy normal?

It is well known that GRF is overly conservative for low degrees of freedom. I am not sure exactly at what df this effect becomes significant but I would not be surprised if you get greater sensitivity with randomise up to ~20-30 subjects.

Also, I would in general discourage the use of a pre-threshold mask and GRF in FSL as the implementation of GRF relies on a there being a large volume/surface ratio of the search volume, something that might not be true for small search volumes. Randomise makes no assumptions about that and will always be valid.

Jesper