This article demonstrates a limitation of the classic Shepard, Hovland and Jenkins (1961) study in which a category structure based on an XOR rule (Type II) is learned faster than a linearly separable category (Type IV). The article documents several instances for which the Type IV category is learned faster than the Type IV category structure. Furthermore, it is shown that when the instructions bias the participants to focus on “rules” and “relevant dimensions”, the Type II advantage re-emerges for some types of stimuli.