Lithification is stressed as a major bias for the palaeobiodiversity evaluation. Although this bias is often discussed in the literature, it has rarely been quantified. This work offers a first estimation of diagenesis impact over mollusc diversity record for a single bed of the “Falunière” of Grignon (middle Lutetian, France). This bed possesses the particularity of displaying two lithological facies: one lithified and the other unlithified, both from a same taphocoenosis. Mollusc diversities of three unlithified and three lithified samples have been compared (1453 specimens among 131 species). The comparison was made possible by the construction of rarefaction curves extrapolated for 30 samples and the introduction of two indexes: the eDG (extrapolated Diagenesis Gap) that gives a value of diversity loss between two facies and the STD (sampling/diagenesis bias threshold) that gives the threshold (in number of samples) after which eDG can be estimated. The analysis reveals that nearly 80% of species richness is not recorded in the lithified facies, and that loss can reach 100% for species smaller than 2 mm. The bias linked to specimen sizes is discussed, both for large and small shells. The differences of biodiversity recorded among litholofacies have also been approached at regional level by the comparison of taxon associations from lithified and unlithified lithologies from the middle Eocene of Paris Basin (Lutetian: Vanves, Nanterre, Damery, Ferme de l'Orme, Chaussy, Grignon and Villiers-Saint-Frédéric; Bartonian: Baron), Aquitaine Basin (Bartonian: Blaye and Gironde) and Italy (Lutetian: San Giovanni Ilarione, Verona). A revaluation of biodiversity estimates of San Giovanni Ilarione that consider lithification bias suggests that the Tethyan regions housed similar or higher species richness than the Paris Basin during the Lutetian, which does not agree with a raw data comparison but which would better fit with the hypothesis of a biodiversity hotspot in the western Tethys. Any future comparisons of the biodiversity from distinct regions or time intervals have to consider the conditions of preservation and the lithification bias.