Although there is a growing body of research on inferential sentences (Declerck 1992, Delahunty 1990, 1995, 2001, Koops 2007, Pusch 2006), most of this research has been on their forms and functions in written discourse. This has left a gap with regards to their range of structural properties and allowed disagreement over their analysis to linger without a conclusive resolution. Most accounts regard the inferential as a type of it-cleft (Declerck 1992, Delahunty 2001, Huddleston and Pullum 2002, Lambrecht 2001), while a few view it as an instance of extraposition (Collins 1991, Schmid 2009). More recently, Pusch’s work in Romance languages proposes the inferential is used as a discourse marker (2006, forthcoming). Based on a corpus study of examples from spoken New Zealand English, the current paper provides a detailed analysis of the formal and discoursal properties of several sub-types of inferentials (positive, negative, as if and like inferentials). We show that despite their apparent formal differences from the prototypical cleft, inferentials are nevertheless best analysed as a type of cleft, though this requires a minor reinterpretation of “cleft construction.” We show how similar the contextualized interpretations of clefts and inferentials are and how these are a function of their lexis and syntax.