Abstract: Existing water pollutant identification methods based on multi-spectral fusion generally have problems such as fuzzy spectral feature contribution degree and weak indicative feature interference, which leads to high decision-making uncertainty in the model. Aiming at this limitation, this paper proposes a method of pollutant identification, which combines multi-spectral contribution rating framework with sequential three-branch decision making. Firstly, a multi-spectral contribution scoring framework based on misclassification cost and information gain was constructed to quantify the differential weights of different spectra for pollutant identification, and the optimal spectral decision sequence was obtained. Then a sequential three-branch decision model is designed to realize the priority decision of spectral features with high differentiation by dynamic threshold division. Through a series of experiments on 5 typical pollutants, the recognition efficiency of this method is evaluated, and the recognition accuracy is 0.83, which is better than other methods.