Hi, Thank you for your speedy response. So all the L2 models in which this issue arises have in common a design comparing copes across different L1 runs. All are specified as fixed effects. Below are some examples: 1) Input images: input 1 (run1, cope4.nii.gz), inputs 2-7 (runs 3-8, cope1.nii.gz) EV matrix: identity matrix where each cope image input is a separate EV (e.g. EV1 = input 1) Contrast matrix: C1, EV1 =-6.0, EVs 2-7 = 1.0 (each) C2, EV1 = 6.0, EVs 2-7 = -1.0 (each) 2) Input images: input 1 (cope1.feat, from L2 average of L1 cope1.feat from runs 3-5) input 2 (cope1.feat, from L2 average of L1 cope1.feat from runs 6-8) EV matrix: EV1 = Input 1, EV2 = Input 2 Contrast matrix: C1, EV1 = 1, EV2 = -1 C2, EV1 = -1, EV2 = 1 *I should note that there is no extreme smoothness in any L2 solely averaging across runs. 3) Input images: inputs 1-6 (cope1.nii.gz, runs 3-8) EV matrix: identity matrix where each cope image input is a separate EV (e.g. EV1 = input 1) Contrast matrix: C1, EVs 1-3 = 1.0 (each), EVs 4-6 = -1.0 (each) C2, EVs 1-3 = -1.0 (each), EVs 4-6 = 1.0 (each) How do these seem? Many thanks, Micah