Abstract: Unsupervised learning with K-means clustering is used to identify pollution characteristics of urban drainage pumping outflow during wet weather. Indicators including pumping station asset property and behavior data are chosen and then profiled for over 200 pumping stations of Shanghai downtown area. It shows that these pumping stations are classified into 4 clusters including low-frequency high-concentration, high-frequency low-concentration, high-frequency high-pollution, and medium-frequency medium-pollution, and the last 2 clusters are of higher priority for pollution control measures. The method used to profile pumping stations shows reasonable results and is of great value for policymakers to deploy drainage quality improving and efficiency enhancing measures.