Clarifying the environmental significance of proxies and establishing a reliable function between regional climate variables and proxies can enhance the accuracy of quantitative precipitation reconstruction, which is benefit for revealing global climate change and their regional environmental responses. However, the links between surface soil phytolith assemblages and precipitation changes in arid and semi-arid regions remain unclear. In this study, we conducted gridded sampling (0.5 degrees x 0.5 degrees) of 41 surface soil samples across a precipitation gradient of 50-400 mm in the arid and semi-arid regions of northern China for phytolith analysis. The results showed that the phytolith morphotypes could be classified into 11 morphotypes, dominated by Gobbet, ELONGATE ENTIRE, and RONDEL. Cluster analysis indicated that the surface soil phytolith assemblages effectively reflect local vegetation. Redundancy analysis (RDA) revealed that, compared to mean annual temperature (MAT), potential evapotranspiration (PET) and wind speed (WS), mean annual precipitation (MAP) is the dominant variable controlling the spatial distribution of phytoliths. CRENATE and BLOCKY were positively correlated with MAP, while Gobbet was negatively correlated with MAP. A new phytolith index PI = (CRENATE + BLOCKY) / (CRENATE + BLOCKY + Gobbet) and a transfer function between the PI index and MAP (MAP = 613.33 x PI +33.36, R2 = 0.8, RMSE = 46 mm) were established. The PI index is more suitable as a proxy for paleoclimate reconstruction in this region compared to previous indices such as Iph. The PI-MAP transfer function demonstrates high accuracy in potential applications in quantitative reconstruction of precipitation in arid and semi-arid regions. This study provides valuable quantitative estimates of paleo-precipitation changes in arid and semi-arid regions.