Comprehension tests often compare accuracy on inferential versus literal questions and find inferential harder than literal, and poor comprehenders performing worse than controls. Difficulties in integration are assumed to be the reason. This research explores another reason--differences in memory for the passage information underlying the questions. Thirty-nine poor comprehenders and 39 controls were given multiple-paragraph passages, which they retold before answering questions. Retellings permitted assessing question accuracy as a function of memory for the text underlying each question. Inferential accuracy was poorer than literal, and the expected group effect was obtained. However, when text memory was perfect, group differences disappeared, indicating that poor comprehenders can generate inferences as well as controls, if they have the relevant information in memory. These findings show that text memory is crucial in distinguishing poor comprehension.