Due to its adverse impact on the human body and environment, air pollution has been an important issue of study, particularly for fine particulate matter (PM2.5). We propose a novel sources and patterns detection technique to analyze the complex physical mechanisms of PM2.5 in central Taiwan. The procedure started with the auto-selecting events mechanism composed of moving average and local extrema calculations, followed by diffusion pattern extraction, which combined the lag-time spatial distribution calculation results from wavelet coherence with principal component analysis to yield the six main diffusion patterns. Finally, representative events were analyzed to discuss the influence of meteorological conditions on the PM2.5 diffusion patterns. The results showed that the general daily PM2.5 concentration variation displayed a bimodal pattern. Among the high PM2.5 events, the cumulative amount was ∼21.35 μg/m3, and the average rise time was ∼8–9 h. Principal Component 1 (PC1) shows the pattern from the coast to inland under the influence of the northeast wind with the highest daily average wind speed (1.56 m/s) and concentration increase percentage (72%); the most serious pollution situation happened in PC5, which is under the influence of weak synoptic, with the highest daily PM2.5 concentration (55.31 μg/m3) and minimum wind speed (1.17 m/s). PM2.5 events with other diffusion patterns (PC2––6) were more likely under the influence of the continental high-pressure peripheral circulation and high-pressure reflux. Overall, the study provides a novel procedure to study environmental problems and a scientific basis for emission control strategies.