Due to the increasing availability of sensors of moderate cost, large scale sensor networks are currently considered for different tasks. One of these is the monitoring of air quality with several, possibly hundreds of sensors that may cover an extensive area, or may be difficult to access. In this paper, we consider the detection of sensor malfunctions in sensor networks. A change in the statistics of a particular sensor can be either due to local variability or a malfunction, and thus simple change detection approaches do not necessarily apply. Here, we estimate the output of each sensor based on all the other sensors using approximate Wiener filters. The approach focuses on the variance of the related prediction errors and also facilitates the control of the trade‐off between the probability of detecting a potential sensor malfunction and the related degree of errors. As a case study, we consider daily ozone data from the Houston air quality monitoring network. Copyright © 2013 John Wiley & Sons, Ltd.