"Best" is a word we tend to avoid, but "good", "reasonable" and "better" are okay. The Euclidean norm (enorm) is usually the one that we use. That's the L2 norm of the backward difference of the motion parameters. The motion parameters are relative to a base volume, not relative to the previous one. afni_proc.py's help has more information under the help for the option "-regress_censor_motion LIMIT" which uses the enorm. You can also see 1d_tool.py help examples 10b, 15a for examples of generating censor lists with enorm or pairwise displacement.
