Measurement of particle size distribution in gas-particle flow is very important. Conventional methods require a complicated classifier like the Andersen Stack Sampler or other particle size measuring instruments. In order to simplify the measurement of particle size distribution, an advanced method employing backward sampling has been proposed. The method measures the particle size distribution without a classifier, and is applicable to melted particles and droplets. There are, however, a few studies regarding the classification efficiency of backward sampling, and the equation that predicts the classification efficiency of backward sampling has not been proposed. In this study, the classification efficiency of backward sampling is numerically and experimentally investigated. Particle trajectories near the sampling probe are calculated to estimate the classification efficiency. Furthermore, experiments with dusts emitted from combustion of pulverlized coal and kerosene are carried out. The concentration ratio (concentration in sampling probe/concentration in main flow) is affected mainly by the particle size, the main flow velocity, and the sampling velocity. The concentration ratio decreases with increasing the main flow velocity and the particle size, but it increases dth the sampling velocity. The concentration ratio is determined by the inertia parameter and the velocity ratio (main flow velocity/sampling velocity). An equation, which estimates the relationship among them, is proposed.