Based on the online real-time interactive multimedia applications (RIMAs) defined in Chap. 1 , this chapter explores the perceptual quality of these applications and our approach to studying these problems. The quality experienced by users is like a black box, with an internal behavior and an input-output relationship that are not well understood. Consequently, users can only perceive the quality through the data received. We explain the concept of perceptual quality and the various function-based evaluation methods. These rely on hypothesizing application-dependent heuristic formulae, whose analytical models are assumed to generalize across different stimuli, with parameters that are then verified empirically. Using a VoIP application as an example, we demonstrate that its perceptual quality cannot be modeled analytically using a function-based approach. We defer to Chap. 4 to present a function-free method for solving this problem. We state the objectives of our optimization process: to minimize the difference in perceptual quality between scenarios with and without network degradations while ensuring that all application requirements are satisfied. The optimization process involves an offline stage to learn single-metric, simplex-control perceptual quality models. In the online stage, the application generalizes the offline models to adapt to current network conditions. It then derives the online multi-metric, multi-control perceptual model for identifying and applying the optimal operating points in the Application Layer.