The objectives of this research are to quantify the magnitude of CO2 exchange and to investigate the responses of net ecosystem CO2 exchange (NEE), gross ecosystem production (GEP) and ecosystem respiration (R-e) to environmental factors over a temperate Stipa krylovii steppe by an eddy covariance system in northern China. Differences in annual precipitation, soil water content (SWC), air temperature (T-a) and evapotranspiration (ET) were the most important factors which controlled patterns and variations of CO2 exchange between atmosphere and the steppe. Precipitation received at study site was 297 mm in 2004, close to the average value during 1971-2000 (+/- S.D., 290 +/- 82 mm), whereas precipitations in 2005 (174 mm) and 2006 (215 mm) were significantly below the average. The lowest values of precipitation and SWC in 2005 resulted in much lower carbon exchange rate related to those in 2004 and 2006. In 2005, daytime NEE reached the seasonal peak about -3 mu mol C m(-2) s(-1), this was much lower than those in 2004 and 2006 (-7 to -8 mu mol C m(-2) s(-1)). seasonal peaks of integrated monthly NEE was the lowest in 2005 (-8.8 g C m(-2) mon(-1)) and the highest in 2004 (-34.3 g C m(-2) mon(-1)). Compared with 2004, the peak values of monthly GEP and R-e in 2005 reduced more than 30% and 20% due to soil water stress. SWC was an important environmental factor regulated the variation of CO2 exchange. At ;hour time scale, daytime NEE was positive when SWC was less than 0.15 m(3) m(-3), R-e tripled when SWC increased from 0.075 to 0.225 m(3) m(-3). At month time scale, SWC was the primary factor controlling monthly CO2 fluxes, there was no net carbon uptake until SWC exceeded 18%, in the low (LAI < 0.5 m(2) m(-2)) and the high (LAI > 0.5 m(2) m(-2)) ranges of LAI, SWC explained about 71% and 90% variations in monthly NEE, and about 92% and 82% variations in monthly GEP. Except SWC, NEE-light response curve was also regulated by air temperature because it influenced both plant photosynthetic capacity and the variations of soil and plant respirations. The optimal air temperature for the carbon uptake of S. krylovii steppe was 10-20 degrees C, carbon uptake decreased in magnitude when air temperature departed from this range. The close relationship between ET and CO2 fluxes was explained that ET was determined by both SWC and T-a. At monthly scale, ET predicted 62% variations of NEE, 78% variations of GEP, 85% variations of R-e. Accumulative ET and accumulative NEE in growing seasons were tightly coupled. The highest ET coincided with carbon sink in the growing season of 2006, and the lowest ET in 2005 resulted in the maximum carbon loss. (C) 2008 Elsevier B.V. All rights reserved.