Abstract: Water use efficiency (WUE) is usually embedded in a variety of ecosystem models to assess the ecosystem response to climate change. However, under natural conditions, multiple environmental factors affect WUE directly and indirectly by influencing the canopy structure. Currently, the mechanisms that influence WUE are not clear. In order to clarify the synergistic effect of various factors on the WUE of winter wheat, experiments were conducted in the Luancheng Agro-Ecosystem Experimental Station, Chinese Academy of Sciences. Variables were observed using an eddy covariance system during key growth stages (greening, jointing, heading, filling) of winter wheat in 2015 (warm and wet year) and 2016 (warm and dry year). The variation in winter wheat WUE and the controlling mechanisms of various factors (net radiation, R_n; air temperature, T_a; vapor pressure deficit, VPD; wind speed, WS; soil water content, SWC) were analyzed by means of a structural equation model (SEM). The structural equation model can systematically analyze the impacts of different factors on WUE on the basis of interactions among different factors. Compared to traditional univariate or multiple linear regression, SEM had intermediate variables, which can decompose the effects of micrometeorological factors into direct and indirect effects. In this study, leaf area index (LAI) was the intermediate variable. The results showed that average WUE in 2015 was 1.52 g(C)·kg^-1(H₂O), while it was 1.22 g(C)·kg^-1(H₂O) in 2016. T_a, LAI, and VPD were the main factors that influenced WUE, regardless of whether the year was warm and wet (WW) or warm and dry (WD). Leaf area index and T_a had positive effects on WUE, while VPD inhibited WUE, which means that under similar temperatures, increased water vapor content in the air can enhance WUE. T_a, LAI, and VPD were of different importance in WW and WD years. LAI was the most significant influencing factor in WW years, while T_a played a more important role in WD years. In WW years, VPD not only affected WUE directly but also indirectly through altering LAI, while it only had a direct effect in WD years. R_n also was different between WW and WD years, having a significant effect on WUE in WW year but no significant effect in WD year. This phenomenon was caused by the heavier and more frequent rainfall in WW year. Obviously, taking the climate conditions in different years into consideration will increase accuracy when simulating WUE. WS had no significant effect on WUE, probably because WS only affects the leaves receive sufficient radiation in the upper part of the canopy, and these effects can be ignored for leaves inside the canopy. Farmland ecosystems have different tolerances and responses to radiation and temperature at different growth stages. LAI can be set as an intermediate variable to reveal this stepwise change in SEM. Therefore, for ecosystems with large seasonal changes in canopy structure, SEM is a powerful tool to investigate mechanisms of environmental control. This research can provide a scientific basis for accurately simulating WUE and predicting the response of WUE to climate change.