I thought the residuals should only change the smoothness estimates, thus shouldn't have on the actual values in the statistics or contrast images. G2 < G3 ( 1 -1 0) is the same as G2 > G3 (1 -1 0). Are you using the same variance settings in both models? Best Regards, Donald McLaren, PhD On Fri, Oct 28, 2016 at 11:00 AM, MRI More <[log in to unmask]> wrote: > In principle you should obtain identical results, independent of the order > of the groups (or more generally, the order of the files loaded). However, > with the default settings SPM calculates residual images for the first 64 > images only. Accordingly, in case your number of subjects is larger (or the > number of files loaded) you will observe some differences. I'm not sure > whether this is really the reason for your observation, but you can try > yourself by setting defaults.stats.maxres in spm_defaults.m to Inf. > > Best regards > > Helmut >