Fish otoliths recovered from pellets of piscivorous birds are often used to reconstruct size and mass of the ingested fish, but erosion of otoliths by gastric fluids usually results in underestimations. We tested the hypothesis that using only uneroded otoliths would greatly improve fish length estimations. In so doing, we compared size-frequency distributions of grayling (Thymallus thymallus) reconstructed from uneroded and eroded otoliths in pellets of Great Cormorants (Phalacrocorax carbo) with size-frequency distributions from a small sample of stomach contents and with samples from various types of fisheries, all from the Rhine River in Switzerland and Germany. Distributions reconstructed from eroded and uneroded otoliths were significantly different. Distributions from worn otoliths failed to delineate different cohorts and produced strong underestimates of fish lengths, thus not identifying larger grayling at all. Distributions reconstructed from otoliths classified as ''uneroded'' represented th whole range of sih sizes eaten and showed the different grayling cohorts, similar to the catch samples, Compared to these, however, average length-at-age in cohort 2 was slightly underestimated (ca. 6-9% in 32 cm long grayling). It is concluded that, at least in fish with relatively large and robust otoliths, careful exclusion of otoliths showing signs of erosion can remove much of the bias in indiscriminately reconstructed size-frequency distributions, thus greatly improving conclusions about the age or size composition of fish in the cormorant diet.