The increasing amount of measurement data on land-atmosphere flux has made it possible to assess the interannual and longer processes that are driven by environmental change and disturbance of terrestrial ecosystems. In this study, I used a global dataset of carbon dioxide (CO₂) fluxes at eddy-covariance tower sites (FLUXNET2015) to investigate long-term trends of net ecosystem CO₂ exchange (NEE), gross primary production (GPP), ecosystem respiration (RE), and related variables. From 118 sites with records of at least 5-years duration, I extracted 1198 site-years of data for use in my analyses. Applying moderate screening by data quality, I found that 58% of the sites showed increasing trends as net CO₂ sinks, in which median slopes of annual NEE of -1.4 and -4.1 g C m^-2 y^-2 were obtained by linear regression and Sen’s slope estimator. Both GPP and RE showed increasing trends at different slopes; their slopes were positively correlated among sites. Across-site variation of NEE trends was analyzed by generalized linear mixed modeling; the best statistical model used temperature, stand age, and biome type as explanatory variables. The trend of increasing CO₂ sinks differed among biome types, from almost none in grassland and savanna to steep slopes in deciduous broad-leaved forest sites. The flux trends derived from terrestrial model simulations showed that the increasing sink trend also prevails over the land. The global model simulations implied that the increasing land sink is primarily attributable to elevated CO₂ concentration. These results demonstrate the usefulness of flux measurement datasets, especially in conjunction with models, to deepen our understanding of long-term terrestrial ecosystem processes.