An integrated approach of principle components analysis (PCA) and Bayesian network (BN) for identify- ing the response pattern of different clusters were developed to understand sensitive relationships of water quality in lakes of Yunnan Plateau. The model includes four steps: data preconditioning, lakes clustering with PCA, Bayesian network learning and lake water quality response modeling. The results demonstrate that the 26 lakes can be clustered into two groups; the Chl a concentration responds more significantly to Total Nitrogen (TN) and Total Phosphorus (TP) in the first group, mainly resulting from much higher watershed disturbances; the Dissolved Oxygen (DO) in the first group with higher water temperature is close to saturation and have little change with Chl a increasing, while the second group is not; and there is good consistency on the relationship between Transpa-rency (SD)and Chl a in both groups.