There are grounds for believing that prompting language learners to infer the meaning of new lexical items is beneficial because inferring the meaning of lexical items and verifying one's inferences invites more cognitive investment than simply being presented with the meanings. However, concerns have been raised over the risk that wrong inferences interfere with later recall of the correct meanings. The present study examines the effect of inferencing on language learners' retention of idiomatic expressions (e.g. "jump the gun," "pull your weight" and "stay the course"). In a counter-balanced within-participant design, 26 advanced learners of English were presented with 21 idioms in contexts either with their meaning cllarified from the start (k = 7) or with the instruction to try and infer their meaning before receiving the clarification. The latter condition was designed so that accurate interpretations were more likely for some idioms (k = 7) than for others (k = 7). The learners' responses at the inferencing stage were collected for analysis. One week later, the participants took an unannounced meaning-recall test. Recall was the most successful in the learning condition where the likelihood of accurate inferences was high. Items that had been inferred accurately stood a better chance (odds ratio 1.22) of being recalled than items whose interpretation had needed to be rectified. Approximately 13% of the wrong or imperfect inferences re-emerged in the post-test, suggesting that the learners did not readily discard them despite the corrective feedback. The findings indicate that, for inferencing procedures to be optimally useful, they need to be implemented in ways that ensure a high success rate.