According to the intelligent complementary strategy, a new watershed water quality assessment method based on rough set (RS) and Bayesian network (BN) was presented for the water quality assessment containing incomplete and uncertain information. Firstly, RS was used to extract the main factors affecting watershed water quality, so as to obtain the minimum attribute reduction set, which can be used to reduce the modelling complexity. Then, the BN was constructed and trained based on the attribute reduction set, and its network structure and conditional probability table were obtained to realize the probabilistic decision reasoning of watershed water quality. Finally, the model evaluation indexes were used to analyse three water quality monitoring sections in the Chongqing section of the Jialing River to verify the correctness and effectiveness of this method. The results show that this method is applicable to the watershed water quality assessment, and has the highest accuracy (>0.97), precision (>0.86), recall (>0.86), and F1-measure (>0.86) compared with other methods (BN, GRA-BN, RS- NB). This method can provide an effective technical support for the water environment management in the watershed.