Free-tagging is one of the leading mechanisms characterizing the so-called web 2.0, enabling users to define collaboratively the meaning of web data for improving their findability. Tagging is applied in several forms to hypermedia data, that represent the largest part of the information on the web. In spite of that, there is a growing part of web data made of 3D vectors representing real objects such as trees, houses and people that lacks any semantic definition. Such situation prevents any advanced use of the data contained inside these 3D worlds, including seeking, filtering and manipulation of the objects represented by vectors. This work proposes a bottom-up approach for adding se- mantics to such data, based on the collaborative effort of users navigating the web. The paper, after describing the similarities and the differences that characterize tagging for hypermedia and interactive 3D worlds, discusses the design choices that have guided the definition of a specification for inserting tags in the latter environments. The approach permits the annotation of most 3D worlds com- pliant with X3D, the ISO standard for describing 3D interactive worlds for the net.