The majority of traditional change-detection tests (CDTs), i.e., techniques for detecting nonstationarity of a data-generating process, are meant for a single data stream. The most straightforward way to perform change detection in a distributed measurement system, where multiple --possibly correlated-- streams are simultaneously acquired by different units, consists in conveying all the measurements to a central processing unit (cluster head) and employ a CDT there. This solution is often unacceptable in wireless sensor networks, which are characterized by serious energy constraints that prevent the continuous data transmission. To this purpose, the change-detection task has to be distributed through the network, exploiting the processing capabilities of each sensing unit. The seminar will present a distributed CDT operating at two-levels: the first, running on each unit, consists in standard CDTs to promptly detect small perturbations in the measurement stream of the unit itself. The second is executed at the cluster head and consists in an aggregated analysis of the network measurements to validate the detection at any unit. According to presented solution, data are not continuously transmitted over the network, because only compressed information must be sent to the cluster head for the validation in the event of a unit detection.