Thank you for your help.
I do not have have a condition with less than 30 TRs, so I should be safe.
However, I realized that the motion censoring option I used is based in the Euclidean norm of the derivative of the motion. So I do not know how a threshold of 0.3 mm compares to my voxel size.
I did an interesting test. I ran the same ANOVA, with and without censoring TRs with too much motion. I got more significant voxels when censoring, so I guess this says that censoring really helps cleaning the signal.
I have another question: censoring TRs based on the derivative of the motion estimators removes spikes due to sudden motions. But what abour slower motion?
Thank you again!
Guillaume