afni.nimh.nih.gov

> after multiple comparison (e.g. FDR) for all 67000 voxels , the p-value of those voxel-clusters are more than 0.05.

I guess you meant the FDR q-value. FDR correction is relatively conservative most of the time.

> I try to use 3dttest++ to do the cluster-level (e.g define 69 voxels as a clusters) multiple comparison
> for my p-values. Can someone help me about it?

Two options: 1) use option -Clustsim, or 2) use option -ETAC. Read the help for more details.

Gang