Terrestrial Biosphere Models (TBMs) are important tools to simulate the carbon cycle of global terrestrial ecosystem, but there are still large uncertainties in their simulation in different regions over the globe. How to evaluate the uncertainty of TBMs is a critical research issue. In this study, taking North America as study area, the uncertainty of gross primary production (GPP) and net ecosystem exchange of CO₂ (NEE) fluxes simulated by CASA-GFED3, SiB3 and SiBCASA was evaluated by using atmospheric concentration data of CO₂ and carbonyl sulfide (OCS) simultaneously obtained from 8 high tower observational stations and WRF-STILT atmospheric particle diffusion model. The results show that the SiB3 model can well simulate the seasonal changes of GPP and NEE in terrestrial ecosystem, and has the best simulation ability among the three models; the CASA-GFED3 model can simulate the seasonal changes of NEE, but there is a large error in the simulation of GPP in growing season; and the SiBCASA model is not able to reasonably simulate NEE and GPP in late winter and early spring. This study proved the role of atmospheric CO₂ and OCS in evaluating the uncertainty of carbon fluxes simulated by TBMs, and provided theoretical support for optimizing the estimation of photosynthetic and respiratory carbon flux of terrestrial ecosystem by using the observation data of atmospheric CO₂ and OCS.