This paper analyzes how speakers can co-encode a reported message and an evaluation of that message in a quotative construction. It presents a typological account of the structures and meanings languages may employ to express, for example, (dis)agreement with or doubt in the truth of the message conveyed and suggests ways in which this may correlate with types of quotative constructions. It argues that interactions between modality and evidentiality in quotatives determine their form and function, and introduces a constructionist model to capture these interactions. By identifying the categories relevant for studying speaker attitudes in quotation, it aims to present a method for the typological analysis of quotatives as ‘double-voiced utterances’, as conceived in Vološinov (1973) and Jakobson (1957).