BACKGROUND: Computerized cognitive training (CCT) has been proposed as a potential therapy for cognitive decline. One of the benefits of CCT is a transfer effect, but its mechanism on the memory domain is unclear. This study aimed to investigate the transfer effect of non-memory multidomain CCT on the memory domain and its neural basis in patients with mild cognitive impairment (MCI) through a randomized controlled trial. METHODS: Patients with MCI recruited from memory clinics were randomly assigned to either the CCT or the control group. The CCT group received multidomain CCT training excluding memory training, while the control group read educational books with learning-based quizzes twice a week for 8 weeks. Participants underwent memory tests yielding a composite score, other cognitive domain tests, non-cognitive scales, and resting-state functional magnetic resonance imaging (rsfMRI), at baseline and after intervention. Within- and between-group comparisons, group × time interactions, and seed-to-voxel analyses in memory-involving brain networks were performed. RESULTS: The CCT group showed improvement over the control group in memory domain (Group × time, F = 5.87, P = 0.03, η = 0.31), which was related with the increased connectivity in the hippocampal-frontal and fusiform-occipital network. No other cognitive and non-cognitive symptoms differed between groups after adjusting for covariates. CONCLUSION: Eight weeks of multidomain CCT without memory training improved memory function and restored functional network in the hippocampal and medial temporal region in MCI patients. These results can provide evidence for the transferring ability of CCT on memory functioning with its neural basis.