Due to human activities, such as the combustion of fossil fuels, the influence on our environment of particulate matter (PM), including aerosols, is being investigated by many research organizations. In particular, fine particles have become very important because they are major components of suspended particulate matter (SPM). Therefore, it is very important to collect fine particles from the exhaust ducts of industrial facilities. Although the method for sampling fine particles less than 2.5 mum (PM2.5) from ambient air has been established by the United States Environmental Protection Agency (EPA), a method for fine particles in exhaust ducts has not been proposed. In this study, a collection method of fine particles by means of backward sampling, in which the sampling probe is aligned at 180 degrees to the main stream velocity, is proposed. This method is investigated numerically and experimentally. The sampling efficiency of backward sampling is calculated by a numerical simulation method under various sampling conditions and an equation, which predicts the sampling efficiency, is derived. The accuracy of this equation is confirmed by measurement of the sampling efficiency in an exhaust duct of a pulverized coal combustion test furnace. By use of this equation, an investigation is made to elucidate the conditions in which the fine particles can be effectively classified.