To formulate effective management strategy and protect the biodiversity of coastal wetlands, there is an urgent need to understand the potential distribution pattern of Spartina alterniflora across different spatial scales. Based on the datasets available of known presences and environmental variables, we set up three study extents from region to global and five grain sizes of environmental variables from 30 arc-second to 10 arc-minute, and used maximum entropy modelling (MaxEnt) to predict the potential distribution of Spartina alterniflora under different spatial scales. On this basis, the habitat areas and its centroid position were measured at three study extents and five grain sizes, respectively. The effects of spatial scale on the relationships between environmental variables and species distributions were analyzed. Results showed that:(1) the performance of MaxEnt model was satisfactory, and the area under ROC curve (AUC) and true skill statistic (TSS) were greater than 0.8 and 0.56, respectively. However, model accuracy was sensitive to changes in spatial scale. AUC and TSS were negative correlated with study extent, but positive correlated with spatial resolution of the environmental variables. (2) There were significant changes in the distribution pattern of Spartina alterniflora across different spatial scales. When increasing study extent from region to global, the suitable area of Spartina alterniflora displayed an enlarging trend, with a significant zonal transfer for geometric centroids, while an increase in spatial resolution of environmental variables reduced the suitable habitats at the regional extent. (3) The importance of main environmental factors would be weakened with the change of spatial scale. At large-scale regions and low-resolution environmental layers, climatic variable type were the main drivers and contributed more than 60% of the variation, whereas at finer scales, this contribution decreased, but that of topography variable increased. (4) Notably, the significant shifts of the prediction accuracy and the distribution pattern occurred when a scale mismatch between study extents and spatial resolution of environmental variables. Therefore, it is recommended to use less than 1.0 arc-minute resolution of environmental variables as input to MaxEnt model for predicting the locally spatial pattern of Spartina alterniflora. When it exceeds 1.0 arc-minutes, the nation or global extent may be more appropriate.