Dear Karsten, Thanks for your kind response. You're probably right that I'm overcomplicating things with this design. However, upon trying a simpler design (i.e. a simple contrast between cert_1 and cert_2) I still have the same problem (see new images attached). In either case, I don't understand why SPM should alter my design matrix when I estimate it. It seems especially strange because I've used the exact same design matrix before with different data and it's worked ok. The only difference with this data (which is source-localised EEG data) is that I performed the inverse solution to the data using SPM, whereas previously I used LORETA-KEY. I don't know enough about SPM to know whether this should make any difference. Any other pointers would be greatly appreciated. Best wishes, Chris