A spatio-temporal filtering method is proposed to detect outliers in wireless sensor networks in this work. Outliers are assumed to be uncorrelated in time and space, and modeled as an alpha-stable distribution. The proposed algorithm consists of collaborative time-series estimation, variogram application, and principle component analysis (PCA). It is realized on self-organized clusters that can manage the data locally. Conceptually, each node detects any temporally abnormal data and transmits the rectified data to a local cluster-head, which detects any survived spatial outliers and determines the faulty sensors accordingly. As a result, faulty sensors do not burden the sink to achieve the following two goals simultaneously, i.e., enhancing the data quality and reducing the communication cost in wireless sensor networks. It is demonstrated that the maximum outlier detection rate is around 94% when the noise level is alpha=0.9.