Prior studies have shown that predictions of subsequent performance (i.e., Judgments of Learning, JoLs) following tests are more accurate than those following re-study and have suggested that retrieval practice allows people to base their predictions on the current retrieval outcomes so that they assign a higher likelihood of remembering to answers with high confidence. We speculated that other mechanisms, such as retrieval fluency during tests, might also be important for JoLs and that they both offer diagnostic information helping learners to make more accurate JoLs. In the present study, we asked participants to study word-pairs and undergo either a test or re-study trial. Two testing formats (cued-recall and multiple-choice) were administrated for the test condition in two experiments. After the initial test or re-study of the word-pair, participants rated their confidence in the current retrieval accuracy (test) or confidence in acquisition (re-study), followed by a JoL rating where participants predicted their performance in the final test one day later. The results of both experiments showed that the correlation between JoL ratings and the final accuracy was higher for test trials compared with re-study trials. Moreover, using mediation analyses, we found that this high correspondence was only partially mediated by participants' confidence in initial tests. Both retrieval reaction time and retrieval confidence simultaneously mediated the correspondence between JoLs and the final accuracy, suggesting that participants were able to correctly base their JoLs on multiple sources of information that are made available through retrieval practice.