This study sought to develop a description of second language fluency, based on the concept that the microcontext of the testing task and the resulting psycholinguistic demands on speech production affect the nature of the discourse produced so that different tasks have differential impact on an individual's display of fluency. It investigated how assessment of learners' fluency varies with four speech task structures (cued dialogue, uncued dialogue, cued monologue, uncued monologue) and any difference in oral discourse produced in performance of the four tasks. Subjects were 46 students in a private school for English as a Second Language in Brazil, divided into three proficiency levels. Each performed one oral task in Portuguese and five in English (two dialogues, two monologues, and a cued monologue from a standardized test), filled out a questionnaire, and was interviewed. Four raters assessed performances on scales for continuity/flow (length of utterance), ease of expression (quality of pause or hesitation), and rate of speech (speed and evenness). As hypothesized, task structure and interactivity both influenced speakers' fluency ratings and discourse variability, with greater fluency apparent in dialogues. Implications for theory of fluency, instructional and testing practice, and research are considered. Contains 61 references. (MSE)