Pollen counting is an important element of research in pollination, and the majority of current methods involve manual counting. An image-processing technique has been developed for accurate counting and sizing of small particles, and this was tested on almond pollen grains. Our automatic process was found to be several times faster than manual counting and more accurate, with an average error of only 3%. Because of its simplicity, it is an attractive tool for research on such topics as pollen viability or germination. It was found that in the methods using vials there were ‘residues’ of c. 17% on the vial walls and on the handling tools. In addition, the first drop from each vial contained, on average, 1·65 times more grains than the average number of grains in all drops, whereas the last drop from each vial contained fewer grains than the average. This phenomenon was correlated with the non-uniform pollen distribution in the suspension. An automatic counting system is therefore, recommended, to avoid the problem, otherwise a calibration procedure should be used that accounts for both the residues on the tools and the non-uniformity of contents of sample drops.