The Holocene moisture–temperature correlation on varying spatial–temporal scales in arid Central Asia (ACA) is still controversial. It has been widely reported that ACA experienced multi-centennial alternations between warm/dry and cold/wet climates over the past two millennia. However, less attention has focused on orbital-scale moisture–temperature relationship at a regional scale across ACA. Here, we contribute a framework including a set of quantitative algorithms to acquire reliable pollen-based climatic reconstructions. We apply this methodology to a new pollen record from a wetland in northern Xinjiang (southeastern ACA) for quantitative reconstructions of moisture availability (actual/potential evapotranspiration ratio, AET/PET) and summer temperature (mean temperature of the warmest month, MTWA) over circa the last 10,300 years. We select AET/PET and MTWA because they are evaluated to be most statistically independent and ecologically significant. The effect of differing spatial extents of calibration-sets on model performance is tested to determine the optimal extent. We critically assess the reliability of all reconstructions through calculations of statistical significance, analogue quality and goodness-of-fit statistics. Our final reconstructions are statistically significant with independent features of AET/PET and MTWA, showing an increasing (declining) trend of Holocene moisture (temperature). This anti-phase pattern is consistent with other records and model simulations across southeastern ACA. The data-model consistency postulates that (i) the glacier meltwater from Tien Shan (‘Water Tower of Central Asia’) is crucial to support major streamflow and watersheds over ACA, and (ii) the Holocene wetting trend may be determined by the interacting effects between decreased summer temperature and increased winter precipitation.