There have been several problems in expermental studies about the mediational processes of discrimination shift learning. Two of these problems are that a non-reversal shift (NRS) learning was treated as a secondary event to be comparred with results in a reversal shift (RS) learning, and that the abilities required in the shifts were only indirectely measured by the amount of subjects' (Ss') verbal knowledge about the relevant dimensions or values. We must devise an effective method to get more direct information about the abilities required in the shifts.