Thus far, grassland ecosystem research has mainly been focused on low-lying grassland areas, whereas research on high-altitude grassland areas, especially on the carbon budget of remote areas like the Qinghai-Tibetan plateau is insufficient. To address this issue, flux of CO2 were measured over an alpine shrubland ecosystem (37°36′N, 101°18′E; 325 above sea level [a. s. l.]) on the Qinghai-Tibetan Plateau, China, for 2 years (2003 and 2004) with the eddy covariance method. The vegetation is dominated by formation Potentilla fruticosa L. The soil is Mol–Cryic Cambisols. To interpret the biotic and abiotic factors that modulate CO2 flux over the course of a year we decomposed net ecosystem CO2 exchange (NEE) into its constituent components, and ecosystem respiration (Reco). Results showed that seasonal trends of annual total biomass and NEE followed closely the change in leaf area index. Integrated NEE were −58.5 and −75.5 g C m−2, respectively, for the 2003 and 2004 years. Carbon uptake was mainly attributed from June, July, August, and September of the growing season. In July, NEE reached seasonal peaks of similar magnitude (4–5 g C m−2 day−1) each of the 2 years. Also, the integrated night-time NEE reached comparable peak values (1.5–2 g C m−2 day−1) in the 2 years of study. Despite the large difference in time between carbon uptake and release (carbon uptake time < release time), the alpine shrubland was carbon sink. This is probably because the ecosystem respiration at our site was confined significantly by low temperature and small biomass and large day/night temperature difference and usually soil moisture was not limiting factor for carbon uptake. In general, Reco was an exponential function of soil temperature, but with season-dependent values of Q10. The temperature-dependent respiration model failed immediately after rain events, when large pulses of Reco were observed. Thus, for this alpine shrubland in Qinghai-Tibetan plateau, the timing of rain events had more impact than the total amount of precipitation on ecosystem Reco and NEE.