Abstract <p>The problem of elemental background variation greatly influences anomaly recognition and has not been effectively addressed. Taking the 1 : 200 000 scale geochemical data of stream sediments from Leiyang, SE China, as an example, we established a Random Forests model to determine Sn anomalies, focusing on the application effect of this method. Al₂O₃, CaO, K₂O, MgO, Na₂O, SiO₂, Ba, Be, Li, and Y were selected as model indicators, and 40 background and 40 anomalous samples were chosen to train the Random Forests model, which was then used to estimate the probability (0 to 1) of Sn mineralization. According to the identification rate of known Sn deposits in the study area, the high, medium, and low Sn mineralization probability thresholds were determined to be 1, 0.98, and 0.96, respectively, and thus, strong, medium, and weak Sn anomalies were delineated. Compared to the Sn anomalies identified using the [mean ± <i>k</i> standard deviation] method and the partitioning method, those determined by Random Forests exhibit greater accuracy. Random Forests eliminated some spurious anomalies that are easily misidentified in high-background areas and better recognized some subtle anomalies that are challenging to detect in low-background areas, suggesting that it can largely remove the background influence on the identification of geochemical anomalies.</p>