Recent clinical trials failed to endorse dichoptic training for amblyopia treatment. Here, we proposed an alternative training strategy that focused on reducing signal threshold contrast in the amblyopic eye under a constant and high noise contrast in the fellow eye (HNC), and compared it to a typical dichoptic strategy that aimed at increasing the tolerable noise contrast in the fellow eye (i.e., TNC strategy). We recruited 16 patients with amblyopia and divided them into two groups. Eight patients in Group 1 received the HNC training, while the other eight patients in Group 2 performed the TNC training first (Phase 1) and then crossed over to the HNC training (Phase 2). We measured contrast sensitivity functions (CSFs) separately in the amblyopic and fellow eyes when the untested eye viewed mean luminance (monocularly unmasked) or noise stimuli (dichoptically masked) before and after training at a particular frequency. The area under the log contrast sensitivity function (AULCSF) of masked and unmasked conditions, and dichoptic gain (the ratio of AULCSF of masked to unmasked condition) were calculated for each eye. We found that both dichoptic training paradigms substantially improved masked CSF, dichoptic gain, and visual acuity in the amblyopic eye. As opposed to the TNC paradigm, the HNC training produced stronger effects on masked CSFs, stereoacuity, dichoptic gain, and visual acuity in the amblyopic eye. Interestingly, the second-phase HNC training in Group 2 also induced further improvement in the masked contrast sensitivity and AULCSF in the amblyopic eye. We concluded that the HNC training strategy was more effective than the TNC training paradigm. Future design for dichoptic training should not only focus on increasing the tolerable noise contrast in the fellow eye but should also "nurture" the amblyopic eye under normal binocular viewing conditions and sustained interocular suppression.