A mobile, closed-chamber system (CC) was used to measure carbon and water fluxes on four land-use types common in the Kazakh steppe ecoregion. Land uses represented crop (wheat or barley, WB), abandoned land (AL), crested wheatgrass (CW), and virgin land (VL). Measurements were conducted during the growing season of 2002 in northern Kazakhstan at three locations (blocks) 15–20 km apart. The CC allowed the measurement of the carbon flux components of net ecosystem exchange (NEE), ecosystem respiration (R_E) and soil respiration (R_S), together with evapotranspiration (ET). Nonlinear regression analyses were used to model gross primary production (GPP) and ET as a function of photosynthetically active radiation (Q); R_E and R_S were modeled based on air (T_air) and soil (T_s) temperature, respectively. GPP, R_E, R_S, and ET were estimated for the entire year with the use of continuous 20-min means of Q, T_air, and T_s. Annual NEE indicated that AL gained 536 g CO₂ · m⁻², WB lost − 191 g CO₂ · m⁻², CW was near equilibrium (− 14 g CO₂ · m⁻²), and VL exhibited considerable carbon accumulation (153 g CO₂ · m⁻²). The lower GPP values of the land-use types dominated by native species (CW and VL) compared to WB and AL were compensated by positive NEE values that were maintained during a longer growing season. As expected, VL and CW allocated a larger proportion of their carbon assimilates belowground. Non–growing-season R_E accounted for about 19% of annual R_E in all land-use types. The results of this landscape-level study suggest that carbon lost by cultivation of VLs is partially being restored when fields are left uncultivated, and that VLs are net sinks of carbon. Estimations of carbon balances have important management implications, such as estimation of ecosystem productivity and carbon credit certification.