Data of historical and ideographic sciences, such as cultural heritage studies, geography, geological evolution, biodiversity, but also experimental data of nomothetic natural sciences, are increasingly documented and published in information systems compatible with predicate-logic that refer to things in reality by unique identifiers (or “keys” most likely to be unique in some context). This can only work as a method of sharing and integrating knowledge beyond the spatial and temporal context of local projects, if the referred features or phenomena of reality are distinct and can diachronically be identified by independent observers in the same way and without a clarifying dialogue between them. In this paper, we argue that only a smaller part of the features in our environment are sufficiently distinct over a useful time-span in order to form such “identifiable individuals”. Ontological categories should each provide specific criteria for the so-called ontological individuation, i.e., about how parts of reality can be subdivided into identifiable individuals that are useful for modelling their behaviour and interactions in reality for answering specific scientific questions, obeying evidential constraints as a result of applying these criteria in observations. We motivate by several examples that there are always cases in which the individuality of such an instance may be undecidable basically within all such ontological categories and that all ontological categories are more or less effective approximations of reality. We argue that effective knowledge sharing by information systems using formal ontologies is only possible if these limitations of applicability and precision to real world phenomena are well understood and taken into account.