This study undertook an investigation of an important problem, so far completely overlooked in the palynological literature — to determine the optimal sample size for pollen grain morphological studies. In other words, we investigated the number of pollen grains which should be measured in order to obtain a representative mean value of a given quantitative feature which, in consequence, would make it possible to more accurately describe the pollen of a given taxon. Investigations were conducted on a sample comprising 3850 Rosa canina L. pollen grains on the basis of the length of the polar axis (P), the equatorial diameter (E) and the P/E ratio, at the flower, specimen and population levels. The size of the pollen samples analysed reflected common sample numbers employed in previous pollen morphology studies, namely from five through 10, 20, 30, 50, 100 up to 200 pollen grains. The statistical analyses performed revealed a relatively low variability in pollen grain biometric features at the levels of flower, specimen and population. At the lowest level of variability analysed, it is sufficient to take measurements of several grains to obtain values satisfactorily representing the variability within the flower level. At the level of a specimen or population, the number of grains necessary to secure representative mean values should range from 15 to 20. However, when the research objective is not only information regarding mean values of pollen grain biometric features but also the analysis of their variability (min—max), then the sample size should include approximately 30 grains. The results obtained, apart from their significance in taxonomic studies, also possess important practical significance; measurements of pollen grain biometric features are very labour-intensive and costly and, sometimes, because of difficulties in obtaining satisfactory quantities of plant material (e.g. herbarium specimens, rare species, paleopalynological collections), also very sparse.