Stand-scale vegetation reconstruction is an important component of pollen-based paleoclimate and paleoenvironment research. Recently, several new quantitative reconstruction models have been reported, the applicability of which needs to be validated before these models can be widely used. In this study, our aim was to assess the applicability of the MARCO POLO model in an arid region (the Qaidam Basin). Top soil pollen collected from 24 plots and surface pollen collected from four lakes were used as "local" and "regional" pollen records, respectively. Our analyses demonstrated that lake size is one of the important factors that influence the reliability of reconstruction. Large lakes in arid regions tend to accumulate over-represented pollen types, leading to higher percentages of these pollen types in lake sediments than in soil samples. The over-represented pollen will probably be identified as regionally derived pollen by the model and incorrectly excluded from the "local" pollen records. Distance between lake and plot can also have a prominent effect. If they are close to each other, lake surface pollen can better reflect the vegetation composition of the plot, which will contribute to obtaining more robust results from present/absence analyses. Pollen representation (R) is a key parameter for the conversion of pollen percentage to vegetation coverage. R can vary considerably among different plots because it is strongly influenced by pollen productivity, dispersal, preservation, and many other external factors. Therefore, it is crucial to estimate R values correctly for the study region. To gain a more comprehensive understanding of the applicability of the MARCO POLO model, further studies need to be conducted in regions with diverse climates or vegetation.