Abstract: A total of 31 suface sediment samples were collected from West Kunlun Mountain in south Xinjiang Autonomous Region in northwest China. These samples are from seven types of vegetation: Picea schrenkiana Fisch. et Mey. forest, Sabina Spach. woodland, sub-alpine steppe, alpine meadow, desert vegetion, cushion-vegetation and vegetation adjancent to glaciers. Pollen percentages and pollen concentrations were calculated in all samples. The dominant pollen types in the region are Chenopodiaceae, Artemisia, Picea, Ephedra, Gramineae, Cyperaceae, Rosaceae, Leguminosae, Compositae etc. In order to reveal the relationship between pollen composition and the vegetation type from which the soil sample was collected, principal component analysis and group average cluster analysis were employed on the pollen data. The results revealed that the major vegetation types in this region could be distinguished by pollen composition: a. Samples from desert vegetation were dominated by pollen of Chenopodiaceae (about 60195%). The percentages of all other pollen types were low. b. Picea forest samples were rich in Picea pollen (about 20%) Sabina forest had more Sabina pollen grains than other vegetation types (about 5%, others <1%). Pollen percentages of Artemisia, Chenopodiaceae and Ephedra were comparatively higher (each about 20%) in these samples from the two types of vegetations. C. Pollen percentages of Artemisia, Cyperaceae, Gramineae and Chenopodiaceae were high in both sub-alpine steppe and alpine meadow. But steppe containal more Artemisia and Chenopodiaceae (steppe 33.75% and 32.30%, meadow 15.57% and 19.48% in average), less Cyperaceae and Gramineae (steppe 2.58% and 7.60%, meadow 22.35% and 12.93% in average) than meadow. d. Samples from cushion-vegetation and vegetation adjacent to glaciers were mainly composed of pollen grains transported from other sites. It was not easy to distinguish them from other vegetation types. Principal component analysis and cluster analysis distinguish samples from Picea forest, Sabina woodland, sub-alpine steppe, alpine meadow and desert vegetation. Therefore we think it will be possible to apply the module to reconstruct past vegetation in this region and other similar regions. Regression analysis was also applied to reveal the relationships between pollen and plant percentages of Artemisia, Chenopodiaceae, Cyperaceae and Gramineae. The results indicated that a linear relationship existed between pollen and plant percentages for Artemisia, Chenopodiaceae and Cyperaeeae.

Key words: Surface pollen analysis, Relationshiop between pollen and vegetation, Pollen assemblage, Numerical analysis, Past vegetation