You should not need to smooth at the first level because the basic GLM is linear. The contrast images are essentially just a weighted combination of the input data: contrast = scan1*weight1 + scan2*weight2 + scan3*weight3 ... + scanN*weightN. If the model was purely linear, it would make no difference if you warp and smooth the individual scans and then add them up, or whether you add them up and then warp and smooth them. However, one thing to note is that the REML estimation of the temporal autocorrelations will be slightly affected by warping and smoothing of the data (partly because different voxels will be selected on which to base the estimation of the covariance). This will change the results slightly (as the weights used to generate the contrast image will changed to accommodate the pre-whitening), but I would not like to say which approach is "best". Best regards, -John On 29 November 2010 13:54, Michel Grothe <[log in to unmask]> wrote: > Dear SPMers, > > the SPM8-DARTEL manual for normalisation of fMRI data states: "In principle (for a random e ffects model), you could run the first level analysis using the native space data of each subject. All you need are the contrast images, which can be warped and smoothed." > Does this mean that I would run the first level analysis on unsmoothed native space data? Or would I also smooth the native space data before first-level analysis and then again the warped contrast images for second level analysis? > I´m using resting state data and the "first-level analysis" would be the creation of individual functional connectivity maps. > > Thanks in advance, > > Michel >