Wireless Sensor Network (WSN) deployment experiences show that data collected is prone to be imprecise and faulty due to internal and external influences, such as battery drain, environmental interference, sensor aging. An early detection of such faults is necessary for the effective operation of the sensor network. We focus on identifying data fault types and their causes. In particular, we propose a hybrid approach to the detection of faults based on three qualitatively different classes of fault detection methods. Rule-based methods leverage domain and expert knowledge to develop heuristic rules for identifying and classifying faults. Estimation methods predict normal sensor behavior by leveraging sensor spatial and temporal correlations, identifying erroneous sensor readings as faults. Finally, learning-based methods are inferred a model for the faulty sensor readings using training data and statistically detect and identify classes of faults. We illustrate the performance of a hybrid approach on data coming from two actual sensor deployments.