The one where you show a bunch of nouns (around 25 I think). Then ask the user to type in as many as they can remember. Then make the one group look at the nouns again after the test, and the control group just left alone.
If his theory is right, the group that studied after the test should do better I think.