This dissertation presents a real-time perceptual rendering framework based on computational visual attention tracking in a virtual environment (VE). The visual attention tracking identifies the most plausibly attended objects using top-down (goal-driven) contexts inferred from a user’s navigation behaviors as well as a conventional bottom-up (feature-driven) saliency map. A human experiment was conducted to evaluate the prediction accuracy of the framework by comparing objects regarded as attended to with human gazes collected with an eye tracker. The experimental results indicate that the accuracy is in the level well supported by human cognition theories. The attention tracking framework, then, is applied to depth-of-field (DOF) rendering and level-of-detail (LOD) management, which are representative techniques to improve perceptual quality and rendering performance, respectively. Prior to applying the attention tracking to DOF rendering, we propose two GPU-based real-time DOF rendering methods, since there have been few methods plausible for interactive VEs. One method extends the previous mipmap-based approach, and the other, the previous layered and scatter approaches. Both DOF rendering methods achieve real-time performance without major artifacts present in previous methods. With the DOF rendering methods, we demonstrate attention-guided DOF rendering and LOD management, which use the depths and the levels of attention of attended objects as focal depths and fidelity levels, respectively. The attention-guided DOF rendering can simulate an interactive lens blur effect without an eye tracker, and the attention-guided LOD management can significantly improve rendering performance with little perceptual degradation.