Two ISO standard methods for in-stack sampling and measurement of PM2.5 and PM10 mass concentrations in flue gas from stationary sources were published in 2009 (ISO 23210, conventional cascade impactors) and 2012 (ISO 13271, virtual impactors). The performances of these two methods in terms of PM2.5 separation efficiency and the accuracy of measured mass concentration were compared at the same sampling point and conditions both using laboratory-scale model flue gas with different dust concentrations and using real flue gas sampled from a test facility for pulverized coal combustion. The virtual impactor showed very satisfactory performance for PM2.5 mass concentration and separation efficiency within the investigated range of mass concentrations and ambient conditions. The conventional cascade impactor method overestimated PM2.5 mass concentration by more than 25% due to particle bounce and re-entrainment of coarse particles from the collection plates. During in-stack PM2.5 sampling from coal combustion flue gas with reactive components at high temperature, the use of greased plates with the conventional impactor caused overestimation of mass concentration, even when grease with high temperature endurance was used. The use of a quartz-fiber filters on the impaction plates reduced overestimation but particle bounce and re-entrainment still remained.