The revelation effect occurs when performing an initial task (vs. no task) increases ratings/endorsements of a subsequent target. I examined whether initial task difficulty (solving an easy vs. hard anagram) moderates the revelation effect in recognition and truth tasks. By a criterion flux account, solving an anagram displaces working memory contents, yielding a liberal response bias. This account predicts a revelation effect only in the recognition task—not influenced by anagram difficulty. By a discrepancy misattribution account, after solving an anagram, discrepantly fluent processing of the target is misattributed to recognition/truth. This account predicts a revelation effect in both tasks that is larger after hard anagrams. The revelation effect on recognition ratings was significant only after hard anagrams, consistent with both accounts, whereas a revelation effect on truth ratings occurred only after easy anagrams, contrary to both accounts. However, I argue that discrepancy misattribution fits best with this unexpected pattern.