This paper presents and evaluates a behavior-based approach to dialogue management, where a system's complete dialogue strategy is viewed as the result of running several dialogue behaviors in parallel leading to an emergent coherent and flexible dialogue behavior. The conducted overhearer evaluation of the behavior-based conversational recommender system CORESONG indicates that the approach can give rise to informative and coherent dialogue; and that a complete dialogue strategy can be modeled as an emergent phenomenon in terms of lower-level autonomous behaviors for the studied class of recommendation dialogue interaction. © 2007 Association for Computational Linguistics.