Changes in emissions and wind are often identified as the two dominant factors contributing to year-to-year variations in the concentration of primary pollutants. However, because changes in wind and emissions are intertwined, it has been difficult to quantitatively differentiate their effects on air quality directly from observed data. In particular, if the annual mean concentration of pollutants is higher than the previous year, it is difficult to identify whether the deterioration in air quality is caused by wind blowing from more polluted regions or an increase in contributing emissions. In this paper, based on wind and pollution roses, we propose a method to differentiate the effects of wind and nonwind (e.g., emissions) changes using direct observation. An index (L) is first defined to quantify the validity of the linear decomposition. The method is then validated by idealized experiments, numerical experiments, and a 2 year observation data set from an actual emissions control program. Finally, we demonstrate the proposed method by studying long-term particulate matter (PM10) variations in Hong Kong during 2000–2011. We find that for most of the period, the linear decomposition of the changes in annual PM10 is valid (up to 90% confidence) and is dominated by the change in nonwind effects (e.g., emissions), whereas the average absolute effect from the wind variability is about 20%. Sensitivity analyses also suggest that our method should work in any location as long as the observed wind and pollution data have sufficient duration and resolution to resolve the corresponding wind and pollution roses.