The agroecosystem plays an important role in global and regional carbon balance due to its large area and high carbon sequestration potential. Due to limitations in observation techniques and the spatial heterogeneity of the environment, net ecosystem CO₂ exchange (NEE) in the rice-wheat rotation system of the Lake Taihu Basin remains poorly understood. This paper aimed to investigate NEE in this agroecosystem based on observations of CO₂ flux with the eddy covariance technique, from December 2011 to November 2013. Annual carbon balance was estimated from observed NEE and carbon contents of grain and straw. Half-hourly flux data were first corrected by removal of anomalous data points, coordinate rotation, frequency loss correction, and WPL correction. Two neural network models for daytime and nighttime NEE gap-filling were built. The results showed that annual daytime and nighttime NEE was -749.49 gC m^-2 a^-1 and -785.38 gC m^-2 a^-1, respectively. Accounting for grain removal and return of straw to the field, total net C absorption was 88.12 gC m^-2 a^-1, characterizing the agroecosystem as a weak carbon sink. This carbon sequestration capacity is greater than those of the Huaihe River Basin, subtropical region, and north China plain. Diurnal NEE exhibited a typical "W" bimodal seasonal pattern, and both seasonal NEE and average monthly diurnal NEE showed significant annual fluctuations. Cumulative diurnal NEE ranged from -12.88 gC m^-2 d^-1 to 5.94 gC m^-2 d^-1, with a mean of -2.10 gC m^-2 d^-1. Maximum cumulative diurnal NEE occurred in the wheat season of each year, with values of -12.88 gC m^-2 d^-1 on April 26, 2012 and -11.62 gC m^-2 d^-1 on April 11, 2013, respectively. Variation in diurnal NEE and daytime NEE in the rice-wheat season was significantly correlated with crop height, suggesting that diurnal NEE and daytime NEE were both influenced by crop growth. Nighttime NEE and soil temperature at 10 cm during the wheat season exhibit a significant exponential relationship, with a temperature sensitivity coefficient (Q₁₀) of 3.03 and 2.67 in 2012-2013, respectively, larger than that of many soil respiration models (Q₁₀ = 2.0). If soil moisture is lower than the field capacity, nighttime NEE is affected mainly by soil temperature; otherwise, nighttime NEE is dually controlled by soil temperature and moisture. There exists a short excitation effect to enhance nighttime NEE after heavy rainfall. Permanent flooding significantly reduces soil respiration, simultaneously decreasing the sensitivity of nighttime NEE to soil temperature. The temperature sensitivity coefficient (Q₁₀) in the rice seasons of 2012-2013 were 1.88 and 1.39, respectively. Alternate change of water regime between the permanent flooding and soil drying in the rice season would generate significant short-term effects of inhibition or excitation on soil respiration, respectively. The effects of carbon flux observation techniques and data gap-filling methods on uncertainty are discussed. Finally, the impact of straw application on greenhouse gas emissions (CO₂, CH₄, N₂O) from the agroecosystem is suggested as an avenue of further investigation.