Hi Stephen, For the 1st level you would code/model the stimuli/responses that are of interest. If I am understanding correctly, you want one for each run, test between training vs. control, not mixing or comparing the runs, but rather, having 3 separate tests, is this correct? If yes, then 3 separate paired t-tests will work. If each run is considered a "modality", then these can be corrected across with "-corrmod" in PALM. If, however, you'd like to compare the three runs, then have a look into the first sheet of this file: https://dl.dropboxusercontent.com/u/ 2785709/outbox/mailinglist/design_karolina.ods This design can be used in FEAT. Hope this helps. If not, please feel free to ask again. All the best, Anderson On 7 February 2017 at 01:28, Stephen Wilson <[log in to unmask]> wrote: > Dear FSL experts, > I am in the process of transitioning to using FSL from another software > package, and I have what is probably a very simple question. I apologize if > this has been answered elsewhere on this listserv; I searched but could not > find anything. I am working through analysis for an fMRI neurofeedback > study. Participants received neurofeedback training (or a control > intervention) across three runs. I expect there to be training effects > across runs, so I would like to set things up so that I do not average > across runs for higher level analyses. What I'd like to end up with are > separate parameter estimates for each run in MNI space, which will allow me > to pull estimates (ideally expressed as percent signal change) for each run > from subject-level ROIs (registered to MNI space) for offline analyses and > to conduct whole-brain analyses comparing, for example, responses during > the third vs. the first run within and across groups. Could you please tell > me how I would need to set up my second level models in this case? Would > it just be a separate EV for each run? Thank you very much for any guidance. >