Global vegetation biomass stores huge amounts of carbon and is thus important to the global carbon budget. For the past few decades, different observation-based estimates and modeling of biomass in above- and below-ground vegetation components have been comprehensively conducted. However, uncertainties still exist, in particular for the simulation of biomass magnitude, tendency, and the response of biomass to natural and anthropogenic drivers. To elucidate these uncertainties, this study compares vegetation biomass of sixteen Earth System Models (ESMs) from the Coupled Model Intercomparison Project Phase 5 (CMIP5) archive with latest observation-based data over the Northern High-Latitudes. Results demonstrate that the models exhibit large variability of vegetation biomass, and the model ensemble mean underestimates temperate forest total biomass but overestimates boreal forest total biomass compared to the observational data. Moreover, both the model outputs and the observational data show individual biomass components are highly sensitive to the change of precipitation across different biomes. Possible causes behind inter-model and model-observation differences, such as the discrepancies of climatic conditions, the carbon allocation schemes, prescribed vegetation distributions, representation of disturbances as well as spin-up processes in the ESMs, are investigated and will be discussed.