Thanks Daniel!
Using the following command, I was basically trying to extract individual subject mean value for each cluster found in group analysis ("anova+tlrc"):
3dROIstats -mask anova+tlrc ${subj}_glm+tlrc > clust.txt
When I initially ran 3dclust on anova+tlrc in order to find all the clusters above a certain threshold 20 clusters were found, 3 of which had only 3 voxels.
However, when I ran the 3dROIstats above, means from those 3-voxel-clusters would not come out and output means only for the rest of 17 clusters (which have more than 3 voxels). So I wondered whether there were minimum number of voxels needed to extract cluster values from each subject?