This paper investigates indirect anaphora (IA) from a corpus-based linguistics (CBL) perspective. Indirect anaphora involves backward-pointing links between surface lexical items, such as demonstrative pronouns and objects, situations or concepts which are mentioned or hinted at in previous discourse, but which are not linguistically encoded as surface grammatical elements such as nouns or noun phrases. This paper reports an empirical study of three types of IA, namely labelling (Francis 1986, 1989, 1994), situation reference (Fraurud 1992, 1992a) and text/discourse deixis (Lyons 1977, Levinson 1983). Although the paper reveals some discernable patterns in the use of IA across different genres, the study ultimately points to the challenges of identifying hard and fast categories of IA in a corpus of real-life examples.