Thinking about the meaning of studied words (deep processing) enhances memory on typical recognition tests, relative to focusing on perceptual features (shallow processing). One explanation for this levels-of-processing effect is that deep processing leads to the encoding of more distinctive representations (i.e., more unique semantic or conceptual features that can be recollected to differentiate the words). This recollective distinctiveness hypothesis predicts that deep processing should reduce false recognition errors, because expecting more distinctive recollections can facilitate retrieval monitoring accuracy (i.e., a distinctiveness heuristic). We report several experiments confirming this prediction, while ruling out explanations based on familiarity or overall memory strength. Additional support for the distinctiveness hypothesis was that a manipulation designed to selectively enhance the distinctiveness of words in the shallow condition eliminated the levels-of-processing effect on false recognition. These findings suggest that conceptual processing can elicit the distinctiveness heuristic, and that recollective distinctiveness drives levels-of-processing effects.