This paper takes a speech-act perspective and examines the degree of validity and reliability of three data collection methods used in speech act research: discourse completion tasks, role plays, and verbal reports. Various formats of these instruments and the type of data they produce under experimental conditions are reviewed and illustrated with examples taken from learners and native speakers. In addition, ways for refining the instrument by focusing on the contextual information of the situation are discussed. This paper shows how role-play data can be analysed sequentially and across multiple turns in controlled settings. It also highlights the significance of using verbal reports as a means of validating experimental data by gaining access to the learners’ cognitive and sociocultural perceptions during speech act performance. It concludes with practical recommendations for refining the instruments used in speech-act research.