Based on 2013-2016 air quality monitoring station data, spatial and temporal characteristics of main modes of PM2.5 in China are analyzed using multivariate data analysis methods including empirical orthogonal decomposition, power spectral analysis and BP canonical correlation analysis. Their correlations with emission and meteorological fields are established. Major conclusions are as follows. There exist two main modes in the PM2.5 field in China. The first is an uniformly increasing mode with the large center located in eastern part of Northwest China and southern part of North China; its time series shows a significant downward trend. The second mode exhibits a dipole distribution with opposite changes in the north and south; the two large centers are located in the central and southern portions of North China and the middle and lower reaches of the Yangtze River, respectively. The first mode of PM2.5 can be regarded as the average state, which is mainly affected by the average emission field, the circulation field and large topography. Impacts of these factors are more significant in the north. The second mode can be regarded as a deviation from the average field, which is more likely associated with cold air activities in the winter. The strength of the cold air determines the location of accumulation of pollutants and the way of transport. Impacts of cold air activities often result in large deviations of pollutants from their averages in the south, and thereby the second mode is more significant in the south. This research effectively utilizes various multivariate data analysis methods to study the evolution mechanism of air pollution in China, which provides scientific and technological supports for further understanding the formation mechanisms for air pollution.