Due to the emergence of smartphones and the ubiquitous wireless network deployment, smartphone user experience has been significantly enriched in recent years. However, the performance of mobile network applications is still far behind our expectation despite the rapid increase in the computational capabilities with more powerful processors and multi-core designs for mobile platforms. One of the fundamental reasons is that many mobile applications are unaware of or do not fully understand the mobile network specific characteristics, leading to inefficient resource utilization, which often results in performance bottlenecks eventually. The goal of my research is to reveal the unique characteristics of cellular networks, evaluate the impact due to such characteristics, and develop performance adaptation techniques accordingly to improve mobile application performance.To achieve the goal, it is essential to dive into all the critical network components along end-to-end network connections including the content partner, the mobile network, the smartphone platform, the application implementation, and the usage pattern. First, the current routing of cellular network traffic is quite restricted, as it must traverse a rather limited number (e.g., 4-8) of infrastructure gateways, which is in sharp contrast to wire-line Internet traffic. Such centralized infrastructure directly affects the network performance, application design, and traffic management. Second, the on-device performance of cellular networks is highly variable, affected by a multitude of factors including channel quality, radio resource allocation, cellular network infrastructure, etc. Therefore, decomposing the impact of individual network components is an essential step towards improving the performance of smartphone applications from the perspectives of users, developers, network operators, and smartphone vendors. Third, unlike desktop applications assuming network resource always desirable, mobile applications have to apply performance adaptation techniques to accommodate performance variation. As cellular networks have very unique characteristics in performance and infrastructure, these techniques have to be aware of such factors. Fourth, as applications are known to be diverse, before we can invent sophisticated performance optimization techniques, a prerequisite is the knowledge of application behaviors and usage patterns. Accordingly, our techniques consist of the following four thrusts: (i) revealing the fundamental infrastructure characteristics of cellular networks that are distinctive from wire-line networks; (ii) identifying the important factors that affect the user perceived performance of mobile network applications; (iii) determining the particular usage patterns of mobile applications; and (iv) developing performance adaptation techniques to optimize the performance of mobile applications accordingly. We will introduce each of the four thrusts respectively in the rest of the statement.