In Experiment 1 (N = 16), under conditions of high memory load (60 pictures and 50 paragraphs) and a 1-week retention interval, undergraduate subjects reported their memory for photographs of scenes (cued recall and free-recall tasks). Subjects frequently reported memory for photographs that they had actually never seen, but had read about in a brief paragraph. In Experiment 2 (N = 40), the same pattern of results was obtained with immediate testing. Experiment 2 also demonstrated that the likelihood of subjects falsely attributing scene memory (based on reading) to actually having viewed a photograph was reduced when metacognitive awareness of imaging during reading was made salient. Awareness of image creation was induced by requiring subjects to rate the paragraphs with respect to imagery vividness. Although other measures of memory remained the same, subjects in the induced-imagery condition made 50% fewer confusion errors than subjects who read the paragraphs without imagery instructions. The results are discussed in the context of Johnson and Raye's (1981) reality monitoring model.