Based on recently developed methods of exploratory spatial data analysis, this paper analyzes the SARS epidemic in China from a spatial perspective. Our results show strong evidence of spatial autocorrelation in the levels of reported SARS cases. A global measure of spatial autocorrelation with Moran's I reveals the dominant pattern of spatial processes being positive, which suggests that provinces with high (low) reported SARS cases tend to be spatially clustered in nature. A local spatial autocorrelation analysis using Moran scatterplot provides a visual impression on the local regimes of spatial dependence as well as the pockets of local non-stationarity with respect to the global spatial process. A further spatial clustering analysis using the local Gi* statistic detects the spatial center of SARS epidemic development and its spatial shifting trajectory. All this strongly suggests that, while regions may be growing or declining in reported SARS cases, they do not do so independently but rather tend to display dynamics similar to their contiguity neighbors.