When the algorithm based on the normal distribution is applied in the sliding window, the detection model will yield an extreme outlier and exhibit unstable phenomenon. To address this issue, this study proposed an algorithm based on the Cauchy distribution for detecting outliers in hydrological time series. Test results indicate that using the median and median absolute deviation in the sliding window instead of the mean and standard deviation can well eliminate the influence of extreme outliers. The setup of the sliding window and the confidence in the algorithm were analyzed. Examples were used to verify this method and the results indicate that the detection rate of the proposed algorithm maintains a high level when the window and confidence settings are appropriate. The comparison with other algorithms shows that the proposed algorithm exhibits good applicability and robustness for dealing with time series with severe local fluctuations.