This study investigated behavioral and cortical mechanisms for short-term postural training with error amplification (EA) feedback in the elderly. Thirty-six elderly subjects (65.7 ± 2.2 years) were grouped (control and EA, n = 18) for training in stabilometer balance under visual guidance. During the training session (8 training rounds of 60 s in Day 2), the EA group received visual feedback that magnified errors to twice the real size, whereas the control group received visual feedback that displayed real errors. Scalp EEG and kinematic data of the stabilometer plate and ankle joint were recorded in the pre-test (Day 1) and post-test (Day 3). The EA group (-46.5 ± 4.7%) exhibited greater post-training error reduction than that of the control group (-27.1 ± 4.0%)(p = 0.020), together with a greater decline in kinematic coupling between the stabilometer plate and ankle joint (EA: -26.6 ± 4.8%, control: 2.3 ± 8.6%, p = 0.023). In contrast to the control group, the EA group manifested greater reductions in mean phase-lag index (PLI) connectivity in the theta (4-7 Hz)(p = 0.011) and alpha (8-12 Hz) (p = 0.027) bands. Only the EA group showed post-training declines in the mean PLI in the theta and alpha bands. Minimal spanning tree analysis revealed that EA-based training led to increases in the diameter (p = 0.002) and average eccentricity (p = 0.004) of the theta band for enhanced performance monitoring and reduction in the leaf fraction (p = 0.030) of the alpha band for postural response with enhanced automaticity. In conclusion, short-term EA training optimizes balance skill, favoring multi-segment coordination for the elderly, which is linked to more sophisticated error monitoring with less attentive control over the stabilometer stance.