Roundness is a useful measure of particle shape which can be used for comparison and distinguishing between different particle classes. Six state-of-the-art automated methods of roundness measurement are presented and compared from the viewpoint of their capability of distinguishing between different roundness classes. The main purpose of this paper is to address the minimal prerequisites of roundness measurement which are required for a successful classification of particle classes, where the roundness is measured using the techniques of 2D image analysis. The first prerequisite concerns the determination of the minimum pixel resolution of analyzed images which contain a projection of analyzed particles. The goal is to determine such a pixel resolution so that the influence of image digitalization was at the minimal possible level. The pixel resolution is obtained based on the correlation coefficient, whose value represents a correlation between referential and automatically measured values of roundness of particles contained in the Krumbein chart. The minimum resolution is determined based on the average length of the projection contour for which the correlation coefficient reaches a saturated value. The second prerequisite is related to the determination of the minimum number of particle projections which is needed for distinguishing a pair of roundness classes. This classification is based on the mean roundness values of the classes. In order to distinguish two classes using random samples, a minimum number of particle projections in the samples are necessary to satisfy that the sample means of the classes will be distinctive with a high probability. Therefore, a procedure is proposed which estimates this minimum number of particle projections for a specific pair of particle classes to be distinguished. The procedure computes the confidence interval of sample means of roundness values and evaluates whether the size of intervals fits to predefined requirements. With this procedure, a relatively accurate minimum number of particle projections can be obtained. Using the determined minimum number of particle projections, a pair of different classes can be reliably distinguished. (C) 2015 Elsevier B.V. All rights reserved.