Recently, researchers in educational psychology pay much more attention to disfluency effect. Studies showed that students’ learning performance was enhanced if materials were presented in a perceptually harder-to-read format, rather than an easier-to-read format (Diemand-Yauman et al., 2011). Previous studies mainly manipulated the extrinsic characteristics of text, picture, and sound in the educationally relevant materials to investigate the role of perceptual fluency in learning. In this article, we firstly reviewed researches about the effects of disfluency on both judgment of learning (JOL) and learning outcomes. Second, theories of disfluency and their potential controversies were introduced and reviewed. Third, based on empirical studies related to perceptual fluency, a meta-analysis was conducted to examine to what extent disfluency would influence students’ learning outcomes as well as JOL. According to previous studies, metacognitive illusions occurred when information was perceptually easy to process, and yielded a sense of overconfidence from fluent materials. However, learning disfluent materials made JOL more cautiously. It’s possibly leading to a remission of overconfidence. However, it remained various mixed results of the effects of disfluency. The generality of the disfluency effect was questioned by many researchers. From the perspectives of desirable difficulty and disfluency theory, perceptual disfluency is likely to improve the storage strength of memory or activate analytical thinking. More subjective efforts and additional cognitive burden are needed while perceptual difficulty is increased. Therefore, learners would process information more deeply, abstractly, and carefully, and result in better retention or comprehension. However, Cognitive Load Theory (CLT) might predict an opposite result of learning for the reason that disfluency bring about an increase of extraneous cognitive load, rather than germane cognitive load. Results of meta-analysis revealed that disfluency lowered learners’ confidence of JOL (d = −0.54, p < 0.01) and hindered learning (d = −0.36, p = 0.06). This result supported Cognitive Load Hypothesis. Finally, implications and further research were discussed. Future studies should focus on the manipulation, the systematic evaluation of disfluency itself, and boundary conditions of disfluency effect.