Virtual environments (VEs) have a set of characteristics that make them hard to be designed and implemented: distributed nature, high-level graphical design, technology novelty, and so forth. Because of the criticism or the repetitiveness of some roles played in them, some of the characters of the VEs usually must be automated. The risk is to pay a too high price, losing attractiveness, usability, or believability. The solution proposed in this chapter is to control the automated avatars by associating them with software agents, becoming intelligent virtual agents (IVAs). With this aim, an architecture to manage the perception and cognition of the agent is described. On one hand, the perceptual module of this architecture consists of a human-like model, based on one of the most successful awareness models in computer-supported cooperative work (CSCW), called the Spatial Model of Interaction (SMI). On the other hand, the cognitive module proposes an easy-to-configure structure, providing it with the precise mechanisms to exhibit reactive, deliberative or, even, more sophisticated social behaviors, incrementing the believability of the IVA in the VE.