Phytoliths serve as a key proxy in paleovegetation reconstruction, offering a valuable methodological approach. However, their morphological redundancy and sensitivity to external environmental factors can introduce bias in representing surface vegetation. Therefore, clarifying and evaluating the preservation characteristics and representation biases of surface soil phytolith assemblages is essential for accurate regional paleovegetation reconstruction. In this study, we collected 34 surface soil and plant community samples from different slope aspects and altitudinal vegetation belts in the Eastern Tianshan Mountains. Following phytolith extraction and identification, we analyzed phytolith types and concentrations, calculated representativeness and preservation indices, and systematically evaluated the preservation characteristics of phytoliths and their capacity to represent plant communities. The results demonstrate that phytolith morphotypes including SADDLE, RONDEL, CRENATE, ELONGATE ENTIRE, ELONGATE GRANULATE, ACUTE BULBOSUS, BULLIFORM and TABULAR IRREGULAR exhibit moderate representativeness and can reliably indicate surface plant communities. Furthermore, using redundancy analysis, correlation analysis, and interpretable machine learning, we quantitatively assessed the influence of environmental factors on phytolith preservation. Elevation strongly influences mountain climate and vegetation development, thereby modifying soil conditions. Among the examined factors, pH and electrical conductivity most significantly affect phytolith preservation. Based on the representativeness of phytoliths across elevational vegetation zones in the Eastern Tianshan, we conclude that phytolith assemblages can effectively indicate vegetation zone evolution along the elevational gradient. However, in piedmont desert and loess depositional areas, paleovegetation reconstruction is challenging due to unfavorable phytolith preservation conditions. In non-zonal piedmont oasis regions, phytolith analysis may offer new insights into oasis evolution processes.