Abstract: Ordination is one of the most popular multivariate analysis methods used in vegetation research. However, most of the commonly used techniques involved in this aspect are concerned with linear data set that is difficult to find in the ecological world. In this paper, non-metric multidimensional scaling,a non-linear ordination technique put forward in Shepard and Kruskal, is introduced and its algorithm is programmed in the C++ computer language as program NMMDS, and it is included in the computer software Gingko V1.O used for structural analysis of vegetation. An example with the classification of tropical montane forest in Bawangling Nature Reserve, Hainan island, is presented. Also other multivariate analysis methods, such as principal components analysis, cluster analysis, are also used for comparison, The result shows that non-metric multidimensional scaling is suitable for higher species richness community classification, and much better than principal components analysis. With this, two sub-vegetation types in Bawangling montane forest have been classified. The tropical montane rain forest is represented by Dacrydium pierrei +Syzygium araiocladum community, while the tropical montane evergreen forest is represented by Sycopsis tutcheri +Syzygium araiocladum community. Also five associations among them are recognized.