Abstract: In this paper, the tropical grassland of the Yinggeling Mountain in Hainan Island was studied using a number of numerical classification and ordination methods, which include two polythetic-hierarchical agglomerative classification (SPSS/PC program)-the nearest neighbour(NN) and the furthest neighbour (FN) methods, the polar ordination (PO) and principal component analysis (PCA) ordination (MINITAB program). The results show that the 19 plots of grassland studied may be classified into 3 types, and 9 plant communities, which are correlative with slope gradient, grazing pressure and soil fertility. It is proved that all the 4 methods are applicable to the study of the tropical grassland. PO is simple and may help clarify those of other methods, PCA and cluster analysis are more objective. PCA and FN are more efficient than FO and NN, and the combination of PCA and cluster analysis can generally produce the most satisfactory result.