In many experiments, memorial benefits have been found when subjects generate items from fragments rather than read items in their complete forms. Does generation cause this difference, or are subjects disposed to adopt different strategies when generating as opposed to reading? If generating causes the difference, items processed in the same way apart from a generative stage should therefore benefit from that generative stage. Our experiments did result in benefits for generating as opposed to reading, but only when the readers processed the words poorly--by pronouncing them. When the readers processed the items well, by imagining them, generating was no better than reading. A new generation effect was found in metamemory. Subjects thought they would remember more generated items than read items; however, the act of making the prediction entailed meaningful processing, and the generated items were not actually remembered better than the read ones.