Driven by interactionist theory and operationalized by task-based interaction, this study aims to investigate EFL learners' task-based negotiation in Second Life (SL), a 3D multi-user virtual environment (MUVE). A group of adult EFL learners with diverse cultural/linguistic backgrounds in L1 participated in this task-based virtual class. Learners used avatars to interact with peers in communication tasks via voice chat. Discourse samples were collected through their oral production to examine their language patterns during negotiated interaction. A framework of negotiation of meaning was employed to code and analyze the transcribed data. Two types of negotiation routine were identified: single-layered trigger-resolution sequence and multi-layered trigger-resolution sequence. Specifically, the interrelationship among task types, negotiation and strategy use was also established in the study: jigsaw task prompted the most instances of negotiation and strategy use, followed by information-gap and decision-making tasks, whereas opinion-exchange task triggered the least. This study suggests that two-way directed tasks with convergent, obligatory, single-outcome conditions will stimulate more cognitive and linguistic processes of negotiation involving interactional modifications -- leading to more complex and lengthy negotiation routine. It is concluded that SL as a 3D MUVE is conducive to theoretically-driven, pedagogically-sound, task-based research in language acquisition.