Abstract The daily average PM2.5 concentrations of Mong Kok (MK), Tsuen Wan (TW) and Tap Mun (TM) in Hong Kong were obtained from Hong Kong Environmental Protection Department and were analyzed. Without any spatial differentiation, PM2.5 concentrations all were higher in winter and lower in summer. In order to more accurately explore the multifractal characteristic of PM2.5 time series, the empirical mode decomposition (EMD) based multifractal detrended fluctuation analysis (MF-DFA) method is used in this paper. The results show that the PM2.5 time series for MK, TW and TM are all anti-persistence, that is, whenever the PM2.5 concentrations has been up (down), it is more likely that it will be down (up) in the close future. The multifractality is caused not only by long-range correlation but also by fat-tailed distribution. Moreover, the roadside site MK displays the highest degree of multifractality, followed by the urban (TW) and the rural (TM) one. The proposed method can more accurately analyze the multifractal characteristic of PM2.5 time series, which provides a solid foundation for further PM2.5 pollution research.