Abstract Among the distinctive features of Occitan varieties spoken in the vicinity of Toulouse is the existence of preterite desinences with a thematic labial (e.g. cantèbi [kanˈtɛβi] ‘sing. prt.1sg ’), or without a thematic consonant at all (e.g. cantèi [kanˈtɛ.i] ‘sing. prt.1sg ’). Such forms contrast with those ordinarily attested in Occitan, where preterite desinences are typically characterised by a thematic rhotic (e.g. cantèri [kanˈtɛɾi] ‘sing. prt.1sg ’), while thematic labials are associated solely with first-conjugation imperfect desinences (e.g. cantavi [kanˈtaβi] ‘sing. ipf.ind.1sg ’). This study draws on data from textual sources and linguistic atlases in order to elucidate the development of the Toulousain labial preterites, which are shown to emerge during the course of the eighteenth century, replacing earlier forms in which preterite desinences systematically presented a thematic velar (e.g. cantègui [kanˈtɛɣi] ‘sing. prt.1sg ’). Since both intervocalic /b/ and /ɡ/ are subject to deletion in Toulousain varieties, it is argued that the development of the labial preterites is not due to direct analogy from the imperfect indicative, but rather to a perception-based change in which speakers miscategorised tokens of /ɡ/ tending to zero as tokens of /b/, and thereby introduced thematic labials into preterite forms. Forms lacking a thematic consonant result directly from deletion of /ɡ/.