Context-aware computing is one of the most essential computing paradigms in pervasive computing. However, current context-aware computing is still in lack of good representation models, particularly in modeling proactive behaviors and historical context data. State diagrams have proven to be an effective modeling method for modeling system behaviors. For context-aware computing, explicitly putting forward states of high-level context can be beneficial and intrigue new angles of understanding and modeling activities. In this dissertation, I propose a state-based context model, and based on the model, I introduce Context State Machines (CSM) for simulating state changes of context attribute, situation, and context, which imply important behaviors of related to context. This research develops and demonstrates CSMs for known context-aware problems from the literature including a smart elevator control system. First of all, the smart elevator, as a context- aware application in the literature, is introduced. Secondly, I introduce the implementation of the CSM engine. Thirdly, I describe two context-aware scenarios, and show the model can help automatically capture the contexts and reason the context without the inference from the developers, and it is the first time in literature to apply state-based modeling approach and the CSM engine to a real-world context-aware system. To evaluate the CSM engine as well as the CSM modeling approach, I generate high-level contextual testing data to feed the engine. I surveyed the data quality issues regarding context- aware software and rubrics of the data quality and dimensionality are developed to address the challenges of applying context to context-aware systems. The rubrics are applied in the generation of synthetic data for feeding the CSM engine in this dissertation.