Based on the eddy covariance observations at 4 sites (2 forested sites and 2 grassland sites) in Chinese Terrestrial Ecosystem Flux Research Network (ChinaFLUX), this paper analyzed the effects of three data processing methods, i.e., spike detection, threshold (u*c) determination of nocturnal friction velocity (u*), and gap-filling model selection, on the estimation of CO2 flux components. All the three methods had significant impacts on the estimation of annual net ecosystem exchange (NEE), and the determination of (u*c) was an important factor affecting the annual NEE estimation. The estimation deviation of the annual NEE caused by spike detection, determination of (u*c), and gap-filling model selection was 0.62-21.31 g C x m(-2) x a(-1) (0.84%-65.31%), 4.06-30.28 g C x m(-2) x a(-1) (3.76%-21.58%), and 0.69-27.73 g C x m(-2) x a(-1) (0.23%-55.62%), respectively. Comparing with that of forested ecosystem, the NEE estimation of grassland ecosystem was more sensitive to the parameter setting of data processing method, and the relative estimation deviation of annual gross ecosystem exchange and ecosystem respiration induced by the uncertainty in data processing was 3.88%-11.41% and 6.45%-24.91%, respectively.