This paper presents a declarative analysis of the phonology of English auxiliaries. The strong and weak forms of auxiliary verbs in English have generally been treated as either related derivationally (Zwicky 1970, Wood 1979, Selkirk 1984) or as lexically suppletive items (Kaisse 1985; this view is also implicit in traditional treatments of English phonetics, e.g. Jones 1960). The derivational treatment involves destructive processes, which Declarative Phonology eschews (Bird 1995, Coleman 1995). The treatment as separate lexical entries fails to address the commonalities observable in related forms such as [hav hbv bv v] for have. This paper provides a declarative analysis of the relations between the multiple forms of English auxiliaries without derivation, and without suppletion. The analysis is based on a corpus as well as data from informants, and is formalised using a computationally tractable formalism. Many of the examples cited in the paper are taken from marsec (Roach et al. 1993), a machine-readable English corpus of material taken from BBC radio broadcasts during the 1980s. The dominant variety of English in marsec is ‘standard’, although in reality this merely means that there is a variety of accents represented which tend towards RP. The database provides natural material rather than idealised or specifically elicited material. As Rischel (1992: 381) notes: ‘Phonology has been based on very exaggerated idealisations about the power of rule machinery as the format in which to take care of variation’. However, some of the structures needed in the analysis presented in this paper do not occur in marsec, so the natural material is complemented by material based on native informants.