Model-based fault detection algorithms can be used to improve the reliability of unmanned aerial vehicles (UAVs) while still satisfying their restrictive size, power, and weight requirements. However, the use of model-based algorithms introduces new failure modes that do not exist in physically redundant architectures. Hence a certification process is needed for such systems that incorporates analysis tools, high fidelity simulations, and ight test data. This paper focuses on one aspect of such a process: the use of ight test data to validate theoretical analysis results. Specifically, this validation is performed to assess the false alarm probability of a simple, model-based UAV fault detection system. This example highlights the main certification issues that arise due to limited ight data and stringent reliability requirements. In addition, the ight test data shows non-Gaussian statistical behavior that leads to some discrepancies with the analysis results. Further discussions are presented for this observed behavior.