AbstractDrought, as an intermittent disturbance of the water cycle, profoundly impacts the terrestrial ecosystem. Recovery time (RT) of an ecosystem from drought is an important indicator of assessing drought impacts and ecosystem resilience, yet the spatiotemporal pattern of ecosystem postdrought RT remains controversial in existing studies. Here we investigate the spatiotemporal pattern of postdrought RT across global terrestrial ecosystems using two observation‐based gross primary productivity (GPP) data sets: direct flux‐site observations and gridded estimates by upscaling flux‐site observations using machine‐learning approach. For droughts that occur on average once every 5.2 years, the RT typically ranges between 2 and 8 months, with a global mean RT of ∼6 months. Spatially, both GPP data sets show a significant bilinear relationship between RT and moisture gradient, and that ecosystems in arid and humid regions tend to recover from drought more rapidly than semi‐arid/sub‐humid ecosystems. Additionally, forests show an overall longer RT than shrublands and grasslands. Temporally, global ecosystem RT shows a slight yet significant increasing trend (0.032 months yr−1) during 1982–2010, which is partly caused by the increasing drought for the same period. However, the observed patterns of RT across global bio‐climatic zones are not captured by the state‐of‐the‐art land surface models, which exhibit a shorter RT in semi‐arid/sub‐humid ecosystems but longer RT in arid/humid regions, and a larger increasing trend of RT over time (0.069 months yr−1). Our findings provide crucial insights into ecosystem vulnerability to sub‐decadal stress events and ecosystem recovery trajectories among diverse bio‐climatic regions and highlight potential model deficiencies that should be accounted for in future model developments.