Ecosystem light use efficiency (LUE) is a critical parameter in estimating CO2 uptake by vegetation from climatological and satellite data. However, the spatiotemporal dynamics and biophysical regulations of ecosystem‐level LUE are not well understood, resulting in large uncertainties in the estimation of gross primary productivity (GPP) using LUE‐based models. In this study, we used eddy covariance to explore spatiotemporal variations and controls of LUE in four contrasting forest ecosystems (savanna, tropical rainforest, subtropical evergreen forest, and subalpine coniferous forest). Based on 27 site years of data, we found that (1) the multiyear mean LUE was 0.063, 0.251, 0.247, and 0.140 g C· mol photon？1 in the four contrasting ecosystems, respectively; (2) the LUE in the wet season (May–October) was higher than that in the dry season in all studied ecosystems; (3) the leaf area index controlled GPP and LUE significantly and explained 74%, 29%, 54%, and 36% of the variation in GPP and 51%, 19%, 41%, and 54% of the variation in LUE in the four contrasting ecosystems, respectively; (4) path analysis revealed the critical roles of GPP and vapor pressure deficit in controlling LUE in these four forest ecosystems; and (5) under warming scenarios, LUE may decrease in savanna but increase in the other three ecosystems, while decreasing precipitation (P) may reduce LUE in the ecosystems studied. This study improves our understanding of the influence of biophysical factors on LUE and demonstrates how LUE changes with variations in temperature, soil moisture, and leaf area index, thereby improving estimations of large‐scale carbon exchange/cycling.