Abstract Vegetation biomass is a key and active component of the carbon cycle. Though China’s vegetation biomass in recent decades has been widely investigated, only two studies have quantitatively assessed its century-scale changes so far and reported totally opposite trends. This study provided the first multi-model estimates of China’s vegetation biomass change for the 20th century and its responses to historical changes in environmental and anthropogenic factors, based on simulations evaluated with the field observations from 3757 inventory plots in China and bias-corrected using machine learning (Gaussian process regression). A significant decline in vegetation biomass over the 20th century was shown by bias-corrected simulations from the six Dynamic Global Vegetation models (DGVMs) with trends ranging from −32.48 to −11.10 Tg C yr–1 and a mean trend of −17.74 Tg C yr–1. Land use and land cover change (LULCC) was primarily responsible for the simulated downward trend (−50.71 to −24.28 Tg C yr–1), while increasing atmospheric CO2 concentration lead to increased vegetation biomass (+9.27 to + 13.37 Tg C yr–1). Climate change had limited impacts on the long-term trend (−3.75 to + 5.06 Tg C yr–1). This study highlights the importance of LULCC for historical reconstruction and future projection of vegetation biomass over China. It also suggests that the incorrect change in China’s forest area for 1980–2000 in the LULCC dataset used as model input data of many existing and ongoing model intercomparison projects (MIPs) has likely led to inaccurate estimations of historical vegetation biomass changes in China.