Based on moderate resolution imaging spectroradiometer (MODIS) 3 km aerosol optical depth (AOD) products, PM2.5 concentration and meteorological data measured near the ground, the spatio-temporal characteristics of AOD and PM2.5 over Xiamen, China, as well as their relationship were studied through comparison of the total and local, unitary and multivariate by using the methods of spatial analysis and statistical regression . The results show there were obvious changes in space-time characteristics of AOD, with highest monthly mean value of 1.133 occurred in April and lowest of 0.635 in January, and seasonal mean value shows a descending trend in spring, summer, autumn and winter, while the annual average value shows a trend of slowly rising, stabilizing and then slightly declining. The higher values of seasonal average AOD were mainly distributed in coastal regions and Xiamen Island, and the lower were mostly located in northwest, north and northeast regions. PM2.5 presented similar spatio-temporal characteristics to AOD, for example, it had obvious seasonal characteristics in time sequence, and in spatial distribution, its higher seasonal mean values were main in central coastal regions, while the lower values appeared mostly in northwest and east regions. There was a moderate positive overall correlation between AOD and PM2.5, with the overall correlation coefficient of unitary linear regression of 0.575. There were obvious differences in the partial coefficients of unitary linear regression between AOD and PM2.5, and the average partial correlation coefficient was 0.432, which was slightly lower than the overall correlation coefficient. The average correlation coefficient of multiplelinear regression of AOD, PM2.5 and meteorological factors was 0.625, and it is shown that the multiple linear regression model of three independent variables has the best fitting effect, but the effect does not tend to be better with the increase of the number of the meteorological factors.