Dear Jian Hou, No, a paired t-test is not appropriate here. And neither is a two-sample t-test. We just want to use the two-sample t-test *design*. This is because we want to create a design matrix with two columns, one for con1 and one for con2. Then we will not do a 1 -1 contrast, as one would for a two-sample t-test. But instead we will do a [1 0; 0 1] F-contrast. This will test for either con1 being non-zero over subjects, either con2 being non-zero, or a linear combination of the two being non-zero. So, this is a bit subtle - we are using the two-sample t-test design, but not doing a two-sample t-test ! All the best, Will. ________________________________ From: jian hou <[log in to unmask]> Sent: 29 April 2016 06:32 To: Penny, William Subject: 2x2x3 model specification Dear William, My experiment is a 2 (group) x 2 (question type) x 3 (valence of picture) design. I searched the spm email list and found your very useful reply regarding model specification and contrasts (https://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=ind1510&L=spm&D=0&1=spm&9=A&J=on&d=No+Match%3BMatch%3BMatches&z=4&P=806176). However, I'm still a bit confused. You wrote that ''To test for (5) the main effect of Factor B (the one with three levels) I would use two contrasts per subject [1 -1 0 1 -1 0] and [0 1 -1 0 1 -1] and take the resulting 70 con images (two per subject) into a two-sample t-test design at the second level. '' I'm wondering if I understand you correctly. Did you mean selecting contrast 1 files (35 images in your example) for group 1 scans and selecting contrast 2 for group 2 scans? Wouldn't a paired t-test be more appropriate since each subject has 2 contrast files? Thank you very much for your help. Best regards, Jian Hou