We conducted field surveys on 20 small, uninhabited islands off the coast of Wenzhou in the Pacific Ocean and recorded 366 vascular plants belonging to 244 genera and 96 families. Of these, 140 were woody and 226 were herbaceous species. Five regression models were constructed to predict species richness (S) and β diversity and the final models were selected based on the Akaike information criterion (AIC). The effects of island size (A), distance to the mainland (I_m), inter-island distance (I_i), shoreline length (Per), perimeter to area ratio (PAR), elevation (E), habitat types (H_n), habitat diversity (H_d), and island shape index (SI) were examined. Model SAH_nR had the highest Akaike weight, 40.26%, with a ratio of 1.19, 2.17, 5.80, and 94.08 over SAR, SAH_dR, BR-SAR1, and BR-SAR2, respectively. Species richness increased with island size following the classic species-area relationship. However, the species similarity index decreased with increasing I_i. The importance of all exploratory variables on β diversity was I_m =0.61, I_i =0.56, PAR=0.49, A= 0.20, P_er =0.14, H_n =0.072, E =0.065, and SI=0.05, suggesting that isolation (i.e., I_m and I_i) played a major role in β diversity. Plant species richness was also significantly affected by PAR and H_d. Stepwise regression showed that the best models to predict S were:S =76.714+1.696A-0.046PAR (R²=0.839), S__woody=6.525+0.455A+24.544H_d (R²=0.697), and S__{herb aceous}=66.899+1.285A-0.04PAR-23.434H_d (R²=0.865) for total, woody, and herbaceous species richness, respectively. This further indicated that species richness of different growth forms (i.e., woody and herbaceous) on these islands was not solely dependent on island size.