Abstract: Online monitoring is an important means of tailings dam warning. How to detect abnormal data that does not match the actual situation of the tailings dam in the monitoring data is the key to improve the accuracy of safety early warning of tailings pond, it is also an important issue in tailings dam monitoring. Three methods including one-class support vector machine, local outlier factor and 3σ rule are used to detect anomalies for three sets of tailings dam monitoring data. Based on classification evaluation matrix, the anomaly detection performance of different methods is studied. The results show that the average precision of one-class support vector machine, local outlier factor and 3σ rule are 0.962, 0.934 and 0.929 respectively; the average recall are 0.960, 0.910 and 0.256 respectively; the average accuracy are 0.984, 0.970 and 0.855 respectively; the average F1 scores are 0.960, 0.921 and 0.393 respectively; the average calculation time are 0.023 s, 7.549 s and 0.348 s respectively. The results indicate that the anomaly detection performance of one-class support vector machine and local outlier factor are significantly better than that of 3σ rule. The anomaly detection performance of one-class support vector machine is better than that of local outlier factor, with a significantly faster calculation speed than local outlier factor as well. The one-class support vector machine is highly accurate, computationally efficient, and performs well in detecting anomalies. The research provides useful reference for tailing pond monitoring warning systems.