The recent development of Brain-Computer Interfaces (BCI) for Virtual Worlds has resulted in a growing interest in realistic visual feedback. In this paper, we investigate the potential role of Virtual Agents in neurofeedback (NF) systems, which constitute an important paradigm for BCI. We discuss the potential of virtual agents to contribute to the success of NF in the specific context of affective BCI. Throughout the paper, we illustrate our presentation with two fully implemented NF prototypes featuring virtual agents. The first is an interactive narrative in which the user empathises with the feature character; the second, an emotion regulation system in which virtual crowd behaviour becomes a metaphor for arousal, as the user attempts to down-regulate their affective state.