Species coexistence mechanisms and species abundance distribution patterns at different time and spatial scale have been and may continue to be key research issues in community ecology, whereas species abundance distribution (SAD) or relative species abundance distribution (RSAD) curve is the common way to describe the species diversity distribution patterns in a community. Species abundance distribution curves have also been used to distinguish different mechanisms for species coexistence. Although there are many different models proposed to predict and explain species abundance distribution patterns and some of them have also been tested in real ecological communities, no general statistical criteria has been proposed to evaluate the power of the models to explain and predict the observed SAD or RSAD. In this paper, we compare the goodness-of-fit indexes of 16 different models for the SADs of plant communities and explore the underlying mechanism of community assembly in sub-alpine meadows in the eastern Qinghai-Tibet Plateau. Taking advantage of Ulrich′s Fortran program for the study of relative abundance distributions, which is named RAD, we fitted 16 different models of species abundance distribution (i.e. geometric model, sugihara fraction model, random fraction model, particulate niche model, etc.) to randomly simulated data sets which were collected in three different types of habitats (north-facing slope, level field, and south-facing slope) by means of two different methods (corrected least square test and chi-square test). We draw the conclusion that more than one model can fit the same data set of the observed species abundance distributions in three habitats. Compared with the other fitted models, the geometric-series model is the best for these three types of habitats in two methods of goodness-of-fit test. The value of fitness index of the geometric-series model always fluctuates around the best fit value 10. In addition to the geometric model, Sugihara fraction model can also fit the observed species abundance distributions in all three habitats but only according to the method of corrected least square test. The goodness-of-fit of other models depend on the types of habitat and the methods of goodness-of-fit test involved. As a result, simple curve-fitting of species abundance distribution might not be treated as the unique criterion for testing the models and their corresponding hypotheses as the mechanisms of community assembly Species traits affecting demographic rate (such as biomass, cover and height) and multiple methods of the goodness-of-fit test should be considered in the comparison of models. Furthermore, searching for a generally optimal model might be one of the future research hot spots because it can play an important role in explaining community structure together with the underlying process in ecological communities. While using multiply models and multiply test methods in our study offers a practicable approach to sift or build a general optimal model and also provides a possibility of producing a unified theory of biodiversity.