Foresight bias is defined in this study as the Judgment of Learning (JoL) made by participants is higher than the correct recall that retrieved by participants. Judgment of Learning is an important form of metacognitve monitoring. It refers to a predictive judgment of whether the items that have learned can be recalled in subsequent test. Previous studies has found that compared to forward associated pairs, backward associated pairs (e.g. rain-umbrella) are more likely to produce higher Judgment of learning. The present study try to examined which ages are more likely to produce higher Judgment of learning with grade 2, grade 3 and grade 5 in prime school, as well as examined which method can reduce the foresight bias with theory-based, mnemonic-based and feedback with the correct recall of items. In the experiment, participants learned a list of 18 pairs backward associated words. Each pair of word finished, participant need do a judgment, which is if you are presented with the word on the left, what the probability that you can recall the word on the right in 5 seconds with a rate of 0-10. After all of the word pairs were finished, participants did a interference task in 1minute, and then start to recall the word on the right with the present of word on the left one by one. Each word were given 20 seconds. After the first “Learn-Test” finished, researcher used different methods to debiasing participants’ foresight bias. In the theory-based group, researcher explained the foresight bias to participants. First, researcher inform the correct recall of items and their predicted to participants. Second, asked participants learned 4 new pairs of word and do judgment, then tell them the real probability they recalled after each word pair. After complete all of 4 word pairs, researcher explained the foresight bias and the potential factors to participants. In the feedback group, researcher just told participants the correct recall and their predicted. In the mnemonic-based group, researcher do not provide any information to participants. Then start the second “Learn-Test” just like the first. The results showed that foresight bias occurred with students of primary 2 and 3, but students of primary 5 did not show the foresight bias. For children of primary 2, only theory-based debiasing procedures had effect on foresight bias, and can lead to transfer to the new materials; for children of primary 3, all three procedures had the same effect for alleviating inflated JoLs. The results showed that matecognitive monitoring is better for older children, and for younger children different debiasing procedures have different effects on reducing their foresight bias.