Stimuli that are processed fluently tend to be regarded as more familiar and are more likely to be classified as old on a recognition test compared with less fluent stimuli. Recently it was shown that the standard relationship between fluency and positive recognition judgments can be reversed if participants are trained that previously studied stimuli are associated with lower levels of fluency. Under such conditions, fluent stimuli are more likely to be classified as new on a recognition test (C. Unkelbach, 2006), which suggests that the interpretation of fluency is malleable and context dependent. Five experiments investigated the resilience of this reversed fluency effect. Using 2 different fluency manipulations, the finding of a reversed fluency effect after training was replicated. However, it was also found that the reversal depends on explicit feedback during training and is specific to the particular fluency manipulation used during training. Moreover, it was found that the reversal did not generalize to similar memory judgments. The balance of experimental results suggest that the standard interpretation of fluency as indicating higher levels of familiarity is quite stable and is resistant to reinterpretation. (Contains 5 tables and 1 footnote.)