Model-based diagnosis is a technique where a model of a system is combined with observations from that system, to generate diagnoses for failures of the system. This thesis looks at how model-based diagnosis can be applied to wireless sensor networks (WSNs). WSNs are ad-hoc wireless networks of small form-factor, embedded nodes with limited memory and processor power. Further, they are often battery powered, meaning that energy use must be kept to a minimum. A diagnoser design is proposed that uses the distributed nature of WSNs to find initial symptoms based on a local model, while leaving the more complex computations required to combine these symptoms to a more powerful central sink computer. A proof-of-concept design is then implemented. Results from this implementation show that using model-based diagnosis in sensor networks is certainly a viable solution. The model used in the proof of concept application created during the work on this thesis did have some problems in dense networks, showing that care must be taken when crafting the model to ensure a successful deployment.