Cyber-Physical Systems are tight integrations of computational and physical processes. Detection of abnormal behavior in cyber-physical systems is crucial for their safe and reliable operation. Model-based detection can improve the system detection performance by incorporating the system behavioral knowledge, captured by the system model, into the design process of detection systems. The components to be modeled, as well as the model complexity, are determined by the specific application and the objective of the study. Classical model-based detection systems have focused on (i) the use of simplified models for the physical system, (ii) the classical abrupt persistent faults, and (iii) the physical system while ignoring other cyber-physical system components. This dissertation addresses the following challenges in model-based detection(i) development of detailed models for physical systems, where simplified models are classically used, (ii) detection of intermittent and incipient faults in physical systems, in addition to persistent faults, and (iii) integration of the wireless communication network in the design of detection systems. These challenges are addressed as follows. First, a novel hybrid dynamical model for aircraft generators is developed to demonstrate the power of physics-based modeling. Second, a general algorithm that is based on change detection theory is developed to detect persistent, intermittent, and incipient faults. The effectiveness of the algorithm is demonstrated by real life case studies. Finally, a unified design process is developed to integrate the wireless communication network and multiple quality measures into the design process of detection systems. The design process is applied on wireless sensor networks with different communication protocols and topologies, and the performance improvement is demonstrated.