Grassland phenology is highly sensitive to climate change. Here, we investigate the spatiotemporal patterns of start (start of season (SOS)) and end (end of season (EOS)) dates of the growing season and quantify changes in their climatic controls over the arid Central Asian grassland ecosystems during 1982-2015, which may improve the model performance by considering shifts in primary drivers under ongoing climate change. Our results suggest that temperature played a positive role in advancing the SOS date, with the control of temperature on SOS getting stronger as preseason conditions become warmer but not drier. For autumn phenology, rapid increase in temperature after 1999 in combination with reductions in precipitation jointly contributed to a shift from delayed to advanced EOS. The areas that EOS regulated by either temperature or precipitation have changed between the two subperiods. Our findings suggest that the dynamic controls of temperature and precipitation on grassland phenology and the difference between spring and autumn phenology should be built into phenological models more accurately. The grassland ecosystems in arid Central Asia are experiencing increased warming and drying. However, the spatiotemporal patterns of grassland phenology and their climate controls are currently unclear. To address this, we employed the empirical orthogonal function to establish the turning points of start of season (SOS) and end of season (EOS) and determined whether the main regulating factors of SOS and EOS dates changed under ongoing climate change. Our results suggest that the SOS and EOS dates both advanced significantly during 1982-2015, with a significant shift in the trends of both SOS and EOS dates before and after the year 1999. The increased advancement in SOS was mainly driven by temperature, whereby the control of temperature on spring phenology became stronger from 1982-1998 to 1999-2015. On the other hand, the EOS was mainly determined by both preseason temperature and precipitation. Lower precipitation combined with increasing temperatures during the period 1999-2015 likely led to a reversal from delay to advancement in EOS. These findings provide insights into the processes of spring and autumn phenology in Central Asian grasslands, which should be implemented in current phenological models to improve our understanding of grassland phenological responses to climate change. The spring and autumn phenology both advanced significantly during 1982-2015, but with a significant shift before and after 1999 The control of temperature on spring phenology getting stronger as preseason conditions become warmer but not drier The shift from a delayed to advanced trends in autumn phenology is related to a combination of warming and reduced precipitation