Disclosed is a method for detecting anomalies in a distribution network equipped with sensors. The steps are: acquiring for each sensor a time series of physical measurements separated by time intervals; defining time windows each corresponding to a plurality of successive time intervals; extracting operational characteristics of each time series in each time window; forming, for each time window, at least one current vector having for coordinates the operational characteristics, structural data relating to the network and situational characteristics specific to the time window; comparing the current vector with preceding vectors corresponding to preceding time windows and the situational characteristics and structural data of which are similar to those of the current vector; and signalling an anomaly in cases where the current vector is significantly dissimilar to the preceding vectors.