Ecological theories and field observations indicate a strong allometric relationship between plant biomass and leaf area. Here, we aimed to rigorously investigate how this allometry can predict the biomass responses to global warming. We conducted a global synthesis on a dataset of 188 species from warming experiments. The reliability of metabolic scaling theory (MST) and functional equilibrium theory (FET) was tested by estimating an allometric coefficient (β) under a Bayesian framework. The results showed that the high β in areas suffering low precipitation was consistent with both theories, while the high β in areas suffering low-temperature stress was consistent with the MST but not the FET. These differences in β between ambient and stressed environments might be derived from the hydraulic constraints in stressed environments. Using general allometry across all species explained 58% of the total variance in the warming responses of plant biomass. The predictive power was not largely improved when factors, such as plant functional type, mean annual temperature and precipitation, warming magnitude, and other experimental treatments, were considered. The predictive error was primarily due to the theoretical assumptions that are based on long-term adaptation failing to capture the changes in specific leaf area (SLA) under rapid global warming. Fortunately, integrating the information on plant traits such as SLA and leaf biomass fraction in the ambient environment effectively improved the predictive power from 58% to 81%, highlighting the necessity of incorporating plant traits into ecosystem models for better predicting the ecosystem carbon cycle in a changing world.