Objective To explore the possible mechanism of compositive cognitive training for promoting cognitive function in the elderly.Methods All the participants were selected from one sub-district in Shanghai.The age ranged from 70 to 75 years old.The subjects were divided into two groups, with 27 in cognitive training group (CT group) , and 22 in control group.The interventions were conducted in small group settings in 24 sessions over 12 weeks.All individuals were assessed with a modified delayed match-tosample (DMS) task to evoke working memory (WM) effects at baseline and post-intervention phase.Results (1) No significant difference form event related potential (ERP) of WM effects between both groups at baseline.(2) After CT, the amplitudes at Fpz, F3, FCz, Cz in P500 component of DMS match to targets task and Fpz, Oz in P500 component of DMS mismatch tasks were higher in CT group than control group.(3) In CT group, the P500 component of match to targets task were significantly different in the amplitude at Fp2, F4, Fpz, and the latency at Fp2, F4, Fz, CPz, Pz of P500 component of mismatch tasks between baseline and post-intervention.The significant difference of change in DMS task was found in the right hemisphere, not in left hemisphere.Conclusions The 3-month compositive cognitive training can improve the cognitive function in community elderly, and these changes are concentrated in the right frontal lobe/prefrontal.