Owing to the strong control bedrock geology may exert on the chemical composition of stream sediments, the determination of stream sediment geochemical anomalies is always affected by the lithology background in areas with variable lithologies. In this study, the expectation–maximization (EM) algorithm was used to separate lithologies of different chemical compositions in a 1: 200 000 scale regional geochemical data set of stream sediments in a lithologically complex region in Hunan province, SE China. The data set included 1024 minerogenic stream sediment samples which were analysed for Cu, La, Li, Be, Cr, Ni, Sr, V, Th, Ti and Zr. A comparison between Cu anomalies determined with and without taking into account the separation of lithologies was carried out. The result shows that stream sediment geochemical anomalies in lithologically complex regions can be determined in a more reasonable way by application of the EM clustering method. Strong but false or meaningless anomalies can be eliminated, and weak but important or meaningful anomalies are more clearly revealed.