Seven undergraduate volunteers studied a written passage on Atomic Structure and then, while answering a set of 24 multiple-choice items, talked aloud about the strategies they were using for option selection. The tape recordings of their verbal responses were analyzed for latency, memory references, and inference references. The items testing knowledge required a shorter time to answer, and the verbal reports contained more words and phrases associated with memory processes, fewer associated with inference, than did those for the items testing higher-order skills. The results suggest the usefulness of a more complex definition of item difficulty. (Author)