The standard treatment of the English preposition of is syntactic by nature; i.e. of is a dummy case-assigner (Chomsky 1981). Contrary to this idea, of imposes strict semantic restrictions on the preceding and following nominals. For instance, in "A of B", A can be a part, and B can be a whole (e.g. the legs of the table), but not vice versa (*the table of the legs). In the product-material relationship, however, both the table of wood and the wood of the table are possible English phrases. This paper aims to spell out ontological properties of of from the perspective of language acquisition. To achieve this goal, I analyzed longitudinal data of eight children at age points of 3, 4, 5, and over 9 from the CHILDES (= Child Language Data Exchange System) database (MacWhinney and Snow 1985, 1990). Using the CLAN software, I extracted all utterances of the subject children that include the uses of of, and analyzed the data with reference to pre-defined sense categories of of. Surprisingly, the well-known examples that are cited as evidence for a dummy case assigner are not found at all in the child language data. Rather, English-speaking children use a handful number of fixed expressions that contain of, most of which are part/quantity phrases or simple collocations. Based on this finding, I argue that of is a relational functor that defines a restricted number of semantic relations over a concept network.