It is not easy to construct a model to describe the geochemical background in geochemical anomaly detection due to the complexity of the geological setting. Isolation forest and its improved algorithms can detect geochemical anomalies without modeling the complex geochemical background. These methods can effectively extract multivariate anomalies from large volume of high-dimensional geochemical data with unknown population distribution. To test the performance of these algorithms in the detection of mineralization-related geochemical anomalies, the isolation forest, extended isolation forest and generalized isolation forest models were established to detect multivariate anomalies from the stream sediment survey data collected in the Wulaga area in Heilongjiang Province. The geochemical anomalies detected by the generalized isolation forest model account for 40% of the study area, and contain 100% of the known gold deposits. The geochemical anomalies detected by the isolation forest model account for 20% of the study area, and contain 71% of the known gold deposits. The geochemical anomalies detected by the extended isolation forest algorithm account for 34% of the study area, and contain 100% of the known gold deposits. Therefore, the isolation forest model, extended isolation fo-rest model and generalized isolation forest model are comparable in geochemical anomaly detection.