Presenting inconsistent postevent information about a witnessed incident typically decreases the accuracy of memory reports concerning that event (the "misinformation effect"). Surprisingly, the "reversibility" of the effect (after an initial occurrence) has remained largely unexplored. Based on a "memory conversion" theoretical framework and associated refined assessment strategy, we report three experiments to demonstrate that suggestive influence can be completely undone. Initially established misinformation effects were eliminated--even after a period of 5 weeks (Exp. 3)--through (a) an "enlightenment" procedure ensuring an adequate representation of the memory task as a search for potentially two contradictory items (instead of "the" single "correct" answer) and (b) using a "memory state test" that unconfounds the performance contributions of item and source memory by assessing them separately. Specifically, memory for original event details that were the target of misinformation was restored to the level of non-misled control performance, and even beyond (Exp. 3). This remarkable reversibility of misinformation influence highlights the central role of memory conversion processes in the misinformation effect (but does not principally exclude the contribution of traditional interference processes). We discuss the compatibility of our findings with previous research and make suggestions for real-world eyewitness interrogation. (Contains 5 tables.)