Despite the rapid advances in computer graphics technology, enhancing the visual quality and lowering the rendering cost is still the essential goal for computer graphics researchers; since improvements in computational power raise the users' expectations too. Especially in interactive 3D games and cinema industry, very realistic graphical contents are desired in real-time. In the meantime, due to the increasing popularity of social networking systems and data sharing, there is a huge amount of data to be visualized effectively. When used carefully, 3D introduces a new data channel for information visualization applications. For that reason, improving the visual quality of 3D computer-generated scenes is still of great interest in the computer graphics and visualization community. In the last decade, utilization of visual perception findings in computer graphics has started to get popular since visual quality is actually judged by the human perception and there is no need to spend additional cost for the physical realism of the details that cannot be perceived by the observer. There is still room for employing the perceptual principles in computer graphics. We contribute to the perceptual computer graphics research in two main aspects: First we propose several perceptual error metrics for evaluating the visual quality of static or animated 3D meshes. Second, we develop a system for ameliorating the perceived depth quality and comprehensibility in 3D visualization applications. A measure for assessing the quality of a 3D mesh is necessary in order to determine whether an operation on the mesh, such as watermarking or compression, affects the perceived quality. The studies on this field are limited when compared to the studies for 2D. A bottom-up approach incorporating both the spatial and temporal components of the low-level human visual system processes is suggested to develop a general-purpose quality metric designed to measure the local distortion visibility on dynamic triangle meshes. In addition, application of crowdsourcing and machine learning methods to implement a novel data-driven error metric for 3D models is also demonstrated. During the visualization of 3D content, using the depth cues selectively to support the design goals and enabling a user to perceive the spatial relationships between the objects are important concerns. In this regard, a framework for selecting proper depth cues and rendering methods providing these cues for the given scene and visualization task is put forward. This framework benefits from fuzzy logic for determining the importance of depth cues and knapsack method for modeling the cost-profit tradeoff between the rendering costs of the methods and their contribution to depth perception. All the proposed methods in this study are validated through formal user experiments and we obtain encouraging results for further research. These results are made publicly available for the benefit of graphics community. In conclusion, we try to make the gap between visual perception and computer graphics fields narrower with the suggested methods in this work.