AIM: To compare the effectiveness of network-based perception learning (NBPL) and traditional training in the treatment of amblyopia children. METHODS: This randomized controlled clinical trial recruited 56 participants aged 4-12y with anisometropic and/or strabismic amblyopia. Participants were randomly divided into two groups: the NBPL group (n=28) who received patching and NBPL for 3mo, and the control group (n=28) who got 3mo of patching and traditional training. Best-corrected visual acuity (BCVA) in the amblyopic eye and stereoacuity were measured and compared at baseline, 1, 2, and 3mo post-randomization. RESULTS: There were no significant differences in age, gender ratio, and BCVA between the two groups at baseline. At 3mo, most patients gained lines (2 logMAR lines on average) of BCVA in both groups except one 11-year-old girl in the control group (P<0.05). But no significant difference in BCVA improvement of the amblyopic eye between the two groups was found (P=0.725), and amblyopia resolved (BCVA of 0.1 logMAR or better or within 1 logMAR line of the fellow eye) for 13 (46.4%) participants in both groups. The number of patients with improvement of stereoacuity was 25 and 13 in the NBPL group and control group (P=0.041), respectively, and a significant difference exists in the distribution of stereopsis at 3mo between the two groups (P=0.015). Besides, in patients with measurable stereopsis improvement degree and space for improvement in the two groups, the NBPL group also achieved better stereoscopic improvement than the control group (10/11 vs 4/11, P<0.05). CONCLUSION: The NBPL system has a significant effect on the improvement of BCVA and stereoacuity of amblyopia children and is better than traditional training in terms of stereoacuity improvement. Perceptual learning visual training may play a more important role in the treatment of amblyopia in the future.