In the first decade of the twenty-first century, we are well and truly at a subject access crossroads, which we have been approaching for some years. From this crossroads three paths diverge. The first is the path of traditional subject access mechanisms, the controlled vocabulary approach represented by lists of subject headings, thesauri and bibliographic classification schemes. The second path leads in the opposite direction: it is represented by derived indexing, that is, the use of terms taken from the content of the information resources itself, rather than terms selected from a controlled vocabulary and assigned by the cataloguer. The third path is the middle way: it advocates the use of the traditional methods of assigning terms from controlled vocabularies augmented by natural language terms derived directly from the information resources, and also, wherever possible, by metadata provided by the sources of the resources – their authors, publishers, and so on. Each of these three paths has its strong and vocal advocates.