The mechanisms behind energy efficiency improvement of coal-fired power generation in China have not been well investigated. We proposed new concepts of learning-by-manufacturing (LBM) and learning-by-operating (LBO) to explain the reduced coal consumption rate of coal-fired power generation. The former results in lower initial coal consumption rates for newly commissioned electric generating units (EGUs) as manufacturers produce better equipment through knowledge accumulation, and the latter leads to a continuous decline in fuel intensity as power plant operators gain experiences in energy conservation. The learning rates of LBM and LBO are estimated at 1.1–1.8% and 1.7–2.1% for different EGUs, respectively. They contributed 662 million tonnes of standard coal equivalent (tce) of energy conservation and correspondingly 1.87 billion tonnes of CO₂ mitigation (36% by LBM and 64% by LBO) during 2000–2017 in China, and can explain 25.8% of the decrease in the national average coal consumption rate.